NBER WORKING PAPER SERIES

INTERGENERATIONAL MOBILITY OF IMMIGRANTS IN THE US OVER TWO CENTURIES

Ran AbramitzkyLeah Platt Boustan

Elisa JácomeSantiago Pérez

Working Paper 26408http://www.nber.org/papers/w26408

NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue

Cambridge, MA 02138October 2019

We thank Chris Becker, Alvaro Calderón, Bruno Caprettini, Raj Chetty, Jennifer Hunt, Michele Rosenberg, Martin Saavedra, Kjell Salvanes, seminar participants at Copenhagen Business School, Copenhagen University, Munich, Princeton, Norwegian School of Economics in Bergen, Rotterdam, Stanford, Tilburg, and UC Davis, as well as conference participants at the 12th International conference on Migration and Development in Madrid, the “Intergenerational Mobility, Gender, and Family Formation in the Long Run” conference in Oslo, and SITE conference on immigration at Stanford for useful comments and discussions. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications.

© 2019 by Ran Abramitzky, Leah Platt Boustan, Elisa Jácome, and Santiago Pérez. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

Intergenerational Mobility of Immigrants in the US over Two CenturiesRan Abramitzky, Leah Platt Boustan, Elisa Jácome, and Santiago PérezNBER Working Paper No. 26408October 2019JEL No. J15,J61,J62,N30

ABSTRACT

Using millions of father-son pairs spanning more than 100 years of US history, we find that children of immigrants from nearly every sending country have higher rates of upward mobility than children of the US-born. Immigrants’ advantage is similar historically and today despite dramatic shifts in sending countries and US immigration policy. In the past, this advantage can be explained by immigrants moving to areas with better prospects for their children and by “under-placement” of the first generation in the income distribution. These findings are consistent with the “American Dream” view that even poorer immigrants can improve their children’s prospects.

Ran AbramitzkyDepartment of EconomicsStanford University579 Jane Stanford WayStanford, CA 94305and NBERranabr@stanford.edu

Leah Platt BoustanPrinceton UniversityIndustrial Relations SectionLouis A. Simpson International Bldg.Princeton, NJ 08544and NBERlboustan@princeton.edu

Elisa JácomePrinceton UniversityIndustrial Relations SectionLouis A. Simpson International Bldg. Princeton, NJ 08544ejacome@princeton.edu

Santiago PérezDepartment of EconomicsUniversity of California at DavisOne Shields AvenueDavis, CA 95616and NBERseperez@ucdavis.edu

A data appendix is available at http://www.nber.org/data-appendix/w26408

2

1. Introduction

Immigrants aspire to offer a better future for their children in the US. Both today and in the

past, many immigrants earn less than US-born workers upon first arrival and do not completely

catch up in a single generation.1 However, a defining feature of the “American Dream” is the view

that even immigrants who come to the United States with few resources and little skills have a real

chance at improving their children’s prospects.

This paper studies the intergenerational mobility of the children of immigrants (the “second-

generation”). We ask whether the sons of immigrants achieve earnings parity with the sons of the

US-born, and how their relative intergenerational mobility changed during the last two centuries.2

On the one hand, children of immigrants might be in a particularly good position to move up the

ladder, given that their parents may earn less than their true talent or ability would suggest (for

example if they had little exposure to US education or if they faced discrimination in the labor

market). Immigrant families who have just arrived in the US might also be more footloose, thus

allowing them to settle in areas with better prospects for their children. On the other hand, children

of immigrants might grow up in segregated neighborhoods, suffer from discrimination themselves

and otherwise enjoy more limited opportunities than the children of the US-born.

We find that, both historically and today, children of immigrants at the bottom of the income

distribution have higher rates of upward mobility than children of the US-born and to a strikingly

similar degree in each time period. Second-generation immigrants growing up at the 25th percentile

of the income distribution end up 3–6 percentile rank points higher than the children of the US-

born. Second-generation immigrants today exhibit a similar degree of upward mobility, despite

notable shifts in countries of origin away from Europe toward Latin America and Asia, as well as

major changes in US immigration policy from a regime of nearly open borders (to European

immigrants) to one of substantial restriction. We find a smaller mobility gap between children of

immigrants and children of the US-born at the top of the income distribution. Second-generation

immigrants growing up at the 75th percentile of the income distribution end up 2–5 percentile rank

points higher than the children of the US born in the past, and less than 2 percentiles higher today.

1 See Abramitzky, Boustan, and Eriksson (2014) for past immigrants and Lubotsky (2007) for more recent immigrants. 2 In historical data, we are only able to link sons from their childhood homes into adulthood across census waves, given that daughters often change their names at marriage. Thus, for consistency, we focus on father-son pairs throughout our analysis.

3

Our analysis focuses on three cohorts of immigrants that span more than one hundred years of

US history. The first two cohorts consist of four million first-generation immigrants observed with

their children in the US in either the 1880 or 1910 Censuses. The 1880 cohort consists mostly of

immigrants from Northern and Western Europe (e.g., Ireland, Germany, and the United Kingdom),

whereas the 1910 cohort includes more immigrants from Southern and Eastern Europe thought to

have faced greater initial disadvantages in the US labor market.3 We follow their children to the

1910 and 1940 Censuses, respectively, using information on their name, year of birth and place of

birth.

The third cohort includes first-generation immigrants observed around 1980 and their children

who are participating in the contemporary labor market. Immigrants in this cohort entered the US

during an era of substantial migration policy restrictions and came mainly from poorer countries

in Latin America and Asia. For this cohort, we use two sources of data. First, we use publicly

available administrative data linking parents and their 2.7 million sons using federal income tax

returns from the Opportunity Insights project (Chetty et al. 2018a, 2018b). Second, we use data

from the General Social Surveys (GSS), which have smaller sample sizes but includes some

undocumented immigrants as well as self-reported information on occupations to facilitate

comparison with the historical data. Note that, because we need sufficient time to observe children

of immigrants in the labor market, we are unable to analyze the most recent cohorts of immigrant

arrivals.

A key limitation of the historical cohorts is that the US Censuses of Population did not collect

income data before 1940. Hence, our analysis of these cohorts relies on computing proxies for

individual-level income (“income scores”). Our preferred income scores use the 1940 Census to

predict income for an individual in earlier Censuses based on his detailed occupation, age,

immigrant status, and location (results are robust to alternative approaches for predicting income,

and for the treatment of farmers’ earnings, which is harder to observe). Our results are qualitatively

similar when we use more contemporary data sources to compute income scores (the 1900 Census

of Agriculture and the 1901 Cost of Living Survey). Immigrants’ mobility advantage is smaller

when we use the occscore variable computed by IPUMS (which assigns occupations their median

1950 income) This finding is not surprising because occscore assigns identical income levels to

3 For more background on immigration during the Age of Mass Migration, see Abramitzky and Boustan (2017).

4

immigrants and the US-born in the same occupation, thereby possibly overestimating immigrant

income and thus underestimating the mobility advantage of their children in the second generation.

We start by documenting (occupation-based) earnings gaps between first- and second-

generation immigrants and US-born workers. In both the past and present, there is a wide variation

in labor market outcomes among first-generation immigrants from different countries. Immigrants

from countries like Finland and Norway in the past and Vietnam and the Dominican Republic

today earn below the US-born on average, whereas immigrants from countries like England

(historical) and India (today) out-earn their US-born counterparts. With a few exceptions, earnings

differences for immigrant groups range between +20 and -40 log points. For most sending

countries in which first-generation immigrants earned less than the US-born, we find that the

second-generation immigrants catch up or even overtake the earnings of the US-born.

Our main analysis uses father-son links to compare the average rank that children can expect

to reach in the national income distribution, conditional on their parent’s income rank and

immigration status (a similar approach to Chetty et al. 2014 and Chetty et al. 2018a). When

analyzing the historical cohorts and the modern GSS data, our ranks are based on occupation-based

earnings, whereas the Opportunity Insights results are based on individual-level income. In both

the past and present, we find that the children of immigrants are more upwardly mobile than the

children of the US-born: conditional on the rank of their parents, children of immigrants have a

higher expected rank in adulthood. The higher level of upward mobility among children of

immigrants is economically meaningful. The difference in outcomes for children of immigrants

with parents in the 25th percentile relative to children of US-born individuals is about one fourth

of the comparable difference between blacks and whites in the historical US (Collins and

Wanamaker 2017), and about one half of the contemporaneous gap between blacks and whites

(Chetty et al. 2018a). The estimated gaps imply that children of immigrants can expect to have

similar outcomes to children of US-born individuals whose parents were ranked about 10–15

percentiles higher in the income distribution.

Importantly, our data also enable us to estimate rates of intergenerational mobility separately

by an individual’s country of origin, both in the historical and the modern cohorts. When doing so,

we find that immigrants at the bottom of the income distribution from nearly every sending

country, including those with a sizable negative earnings gap in the first generation, have higher

5

rates of upward mobility than the children of the US-born. This finding stands in contrast to the

view that immigrants of certain countries of origin are not be able to integrate into the US economy.

In the last part of the paper, we explore the question of why children of immigrants were more

upwardly mobile, focusing primarily on the historical data but comparing with the modern data

whenever possible. We are more limited in our ability to investigate mechanisms in the modern

data since the Opportunity Insights data are available only in aggregate form and the GSS has

small sample sizes and contains limited information about parental background). First, we explore

whether the higher observed mobility of the children of immigrants can be explained by the fact

that their fathers’ rank in the US income distribution did not fully reflect their ability or earnings

potential (for example, due to a lack of US-specific human capital, English language ability, or

labor market networks). Consistent with this explanation, we find the highest rates of upward

mobility for children whose fathers immigrated to the US as adults, compared to those who arrived

as children (and hence were in a better position to acquire US-specific human capital). Father’s

age of arrival is particularly important for immigrants from non-English-speaking countries,

suggesting that fathers’ lack of English ability is one reason why immigrant fathers’ rank does not

fully reflect their earnings potential.

Second, we consider the possibility that immigrant families invest more in the upward mobility

of their children, either by encouraging educational attainment or by moving to geographic areas

with higher mobility prospects for their children. Education does not explain the higher income

mobility of second-generation immigrants. The children of immigrants were not more educated

than the children of US-born of comparable family income (although it is important to note that

immigrant fathers had lower literacy rates at every income level, suggesting higher rates of

educational mobility within immigrant families). Geographic choices were important: first-

generation immigrants were more likely to settle in areas with higher mobility prospects for their

children. When we compare children growing up in the same US region, the intergenerational gap

between immigrants and the US-born is reduced by 70%. When comparing children growing up

in the same county, we no longer find an intergenerational gap between the children of immigrants

and US-born individuals. In other words, immigrant children did not earn more than others who

grew up in the same location. Rather, their parents chose to live in locations that offered high

mobility prospects to all.

6

Our paper contributes to the understanding of immigrant assimilation and intergenerational

mobility in the US more broadly. Most existing work on economic outcomes during the Age of

Mass Migration focus on assimilation within a generation (see for instance, Abramitzky, Boustan,

and Eriksson 2014; Ferrie 1997; Hatton 1997; and Minns 2000). Although immigrants do not fully

catch up to the US-born within one generation, we document substantial progress for the second

generation. Most closely related to our study is Ward (Forthcoming), which documents persistent

outcomes by country of origin in the early 20th century even after controlling for own father and

grandfather attainment.4 We go beyond Ward by contrasting social mobility rates for the children

of immigrants and the US-born and by comparing mobility rates in both the past and present. Our

work also relates to a set of papers that focus on intergenerational progress of specific immigrant

groups, including the Irish famine immigrants who arrived in the US in the mid-19th century

(Collins and Zimran 2019), and Mexican-Americans in the early to mid-20th century (Kosack and

Ward 2018; Duncan et al. 2017).

Second, we add to the broader empirical literature on intergenerational mobility in the US

which, to date, has not emphasized immigrant families. A number of papers estimate contemporary

levels of intergenerational mobility in the United States (e.g., Chetty et al. 2014, 2017; Davis and

Mazumder 2017; Hilger 2016; Lee and Solon 2009; Mazumder 2015).5 A series of related studies

document historical rates of intergenerational mobility, and compare these historical rates to

present-day levels (e.g., Card, Domnisoru, and Taylor 2018; Derenoncourt 2019; Feigenbaum

2015, 2018; Ferrie et al 2016; Long and Ferrie 2013; Olivetti and Paserman 2015; Olivetti,

Paserman, and Salisbury 2018; Tan 2018). Borjas (1992) develops a theory of ethnic capital, which

suggests that the social mobility process may differ between immigrants and the US-born.6

2. Data

A. Historical Datasets for Studying Economic Mobility

4 In earlier papers, Borjas (1993) and Card, DiNardo, and Estes (2000) measured the average outcomes by country of origin in the first and the second generations from Census cross-sections. Perlmann and Waldinger (1997) and Perlmann (2005) compare the outcomes of second-generation immigrants in the past and the present (similar as in our Figure 1) but do not have parent-child linkages to examine intergenerational mobility. Kasinitz, et al. (2009) offers a detailed qualitative and quantitative study of second-generation children in New York City in the current period. 5 Bhattacharya and Mazumder (2011), Chetty et al. (2018a) and Collins and Wanamaker (2017) focus on black-white differences in mobility, and Hilger (2017) focuses on Asian Americans. 6 Outside of the US, Aydemir, Chen, and Corak (2009) focus on the intergenerational mobility of immigrants to Canada, Bolotnyy and Bratu (2018) on immigrants and refugees to Sweden, and Gielen and Webbink (2019) on immigrants from Surinam to the Netherlands.

7

Our main analysis estimates intergenerational mobility for the children of US-born parents and

children of immigrants from 1880 to 2015. We measure historical mobility rates using two new

datasets of linked Census records. The first dataset links children observed in the 1880 Census of

Population to the 1910 Census, and the second links children observed in the 1910 Census to the

1940 Census.7 These data allow us to observe an individual’s own labor market outcomes during

adulthood and his father’s labor market outcomes during his childhood. Throughout the analysis,

we focus on men; women cannot be systematically matched across Censuses because they

typically change their last names after marriage. We define immigrant status based on the father’s

place of birth, but Section 6 shows that the results are similar when defining immigrant status using

the mother’s place of birth (or using both parents’ birthplaces in the historical data when such

information is available).

i. Linking Fathers and Sons

To create each linked sample, we first identified all males aged 0–16 in the childhood Census

(that is, either in 1880 or in 1910). We then matched these individuals to a later Census using

information on first and last names, age, and state of birth.8 Our baseline samples use the linking

algorithm in Abramitzky, Boustan, and Eriksson (2014), although in a later section we show that

our findings are robust to alternative linking procedures.9 We restrict the analysis to father-son

pairs in which: (1) both the father and the son were white,10 (2) the son was living with his father

7 For reasons of both data availability and historical interest, our analysis does not include immigrants in the 1940–1970 cohorts. First, the 1940 Census is the last Census for which information on names is publicly available, allowing for backward but not forward links from 1940. Second, neither of the two sources that we use to study the modern period include immigrants in these mid-20th-century cohorts: the GSS started collecting data in 1972 and the Opportunity Insights data covers children of immigrants born between 1978 and 1983. We also note that these cohorts coincide with a period of relatively limited migration into the US: in 1970, only 4.7% of the US population was born abroad (compared to nearly 15% in the past [1880–1920] and today). 8 The fact that the linking is based on names could be problematic if the likelihood of an individual changing his name was correlated with rates of intergenerational upward mobility. We find this argument unlikely for two reasons. First, all matched individuals are born in the US and so do not go through the name Americanization process that often took place at Ellis Island for immigrant arrivals. Second, we show that names of second-generation immigrants were very unlikely to change over time. To do so, we use the cross-sectional data from 1880, 1910 and 1940 and show that the proportion of a cohort with a given name is very stable across census years. Moreover, names of second-generation immigrants do not appear to become less foreign-sounding in later census years. We provide further details on this issue in Section 1 of the Online Appendix. 9 For more details on the linking strategy, we refer the reader to Section 2 of the Online Appendix. An extended comparison of linking algorithms can be found in Abramitzky et al. (2019). 10 More than 95 percent of immigrants in this time period were white. Focusing on US-born whites as the comparison group ensures that the higher mobility of second-generation immigrants that we observe is not driven by black-white differences in intergenerational mobility. When including African Americans in the sample, immigrant mobility advantage is even higher.

8

at the time of the earliest Census (so that we can observe a father’s labor market outcomes),11 (3)

the father was born in the US or in one of the most common immigrant sending countries during

this time period,12 and (4) both the father and the son were aged 30–50 by the time we measure

their labor market outcomes.13 Finally, we exclude father-son pairs with missing information on

occupation or income, although we show that our main findings are similar when we assign an

income of zero to those with missing occupation or income (see Appendix Figure A22).

Appendix Table A1 shows the changing sample size of the cohorts as we impose each

restriction. In our baseline samples, we match 23% of individuals in the 1880 Census to 1910 and

29% of individuals in the 1910 Census to 1940, standard match rates for historical samples given

the presence of common names and various causes of non-matches (mortality, return migration,

under-enumeration, transcription error; see Abramitzky et al. 2019). After imposing the sample

restrictions, our baseline sample of linked men in the 1880–1910 cohort includes around 1.2

million men, of whom 28 percent are classified as second-generation immigrants. Germany,

Ireland, and England are the largest countries of origin for immigrant fathers in the sample. The

1910–1940 cohort has information on roughly 2.7 million individuals, 24 percent of whom are

sons of immigrants.14 The largest countries of origin in this cohort are Germany, Russia, and

Canada. Appendix Table A2 shows the sample sizes for each of the cohorts, disaggregated by the

father’s place of birth.

An important concern with Census links based on the available information on names, age,

and place of birth is the potential for linking children to the wrong adults (“false positives”). As

our main focus is on differences between immigrants and the US-born, false positives are a concern

only if our linking algorithm is systematically less accurate for the children of immigrants than for

children of US-born individuals (or vice versa). Such a pattern might arise, for instance, if

11 Most children in this age group were living with their father regardless of immigration status. 88 and 90 percent of children with a US-born father lived with their father in the 1880 and 1910 Census, respectively, which is similar to 89 and 90 percent of children with a foreign-born father. 12 The seventeen source countries we consider are: Austria, Belgium, Canada, Denmark, England, Finland, France, Germany, Ireland, Italy, Norway, Portugal, Russia, Scotland, Sweden, Switzerland, and Wales. These countries are the largest source of immigrants in the 1850–1910 period, accounting for 93 and 88 percent of the foreign-born population in the 1880 and 1910 Censuses, respectively. We restrict the analysis to the relatively large sending countries so that we have enough observations to estimate our country-by-country results. 13 We restrict our sample to men ages 30–50 in order to measure fathers and sons at the same moment in the earnings lifecycle, and to avoid measuring income when children or parents are too young or old (Mazumder 2005). 14 Even though roughly 14 percent of the population was foreign-born in 1880 and 1910, immigrants were over-represented in the prime-age adult population. For example, among adult men ages 30–50 who have a child present in the household, 31 and 27 percent of them are foreign-born in the 1880 and 1910 Censuses, respectively.

9

immigrants had names that were less culturally familiar and thus harder for Census enumerators

to transcribe. Linking a child to the wrong adult would lead to attenuation bias in the relationship

between childhood background and adult outcomes; these errors would tend to bias all adults to

the 50th percentile (the population average), thus overstating upward mobility for children whose

parents were at the bottom of the income distribution and overstating downward mobility for

children whose parents were at the top. In the robustness section of the paper, we show that our

results are robust to the use of more conservative linking algorithms designed to minimize the rate

of false positives (at the expense of trading off on sample size).

A second concern with the linking procedure is the extent to which we are able to generate

representative samples. It is well-known that linked samples are not fully representative of the

population, as individuals with uncommon names and other attributes are more likely to match

between Census waves (Abramitzky et al. 2019).15 Appendix Table A3 compares the sons in our

linked samples to the relevant population of US-born men in the 1910 and 1940 Censuses.16 Our

1880–1910 sample of linked sons has roughly the same share of second-generation immigrants as

the full population (26 vs. 27 percent), and has an income score (as described in the next section)

that is roughly 3 percent higher than the population. Our 1910–1940 linked sample has fewer

second-generation immigrants than the full population (22 vs. 24 percent), and again has an income

score that is 5 percent higher than the population. We address the non-representativeness of our

linked data by running a weighted version of our main specification that matches the population

on observable attributes.

ii. Assigning Income Measures in Historical Data

Because the 1940 Census is the first US Census to include information on individual earnings,

we need to construct proxies of individual income for fathers and sons in our historical samples

(we sometimes refer to these proxies as “income scores”). Our preferred approach is to use a

statistical model to predict income from a rich set of covariates for white men ages 30–50 in non-

agricultural occupations in the 1940 Census and use this model to predict income for men in earlier

years. In particular, we regress log income in 1940 on a number of fixed effects and complete set

15 Appendix Table A2 shows the algorithm’s match rate by the father’s country of origin. The match rate for second-generation immigrant sons tends to be slightly lower than for sons of US-born fathers. 16 That is, we focus on white men aged 30 to 50 whose fathers were either born in the US or in one of the source countries of interest.

10

of interaction terms using 3-digit occupation, age, current state of residence, and country of origin

as our explanatory variables (in all interaction terms, we interact covariates with Census region,

instead of state). This method for assigning income is similar to the machine-learning approach

for calculating occupational scores proposed by Saavedra and Twinam (2018).17

The 1940 income variable excludes income from self-employment and hence it is missing or

zero for most farmers. We thus use a different approach to predict the earnings of farm owners,

managers, and laborers. Following Collins and Wanamaker (2017, hereafter CW), we multiply the

known income of farm laborers in 1940 by the ratio of earnings for farmers versus farm laborers

in the 1960 Census by region and immigration status to impute the income of farmers in 1940 (the

1950 Census only asked the income question to a small subset of the population and so it is not

useful for this exercise).18 This measure places the average farmer father in the 40th and 36th

percentile of the income distribution in 1880 and 1910, respectively, although this placement varies

by immigrant status.19

We combine predicted income for non-agricultural occupations with these farming

adjustments to create our preferred “income score.” One concern with this strategy is that there is

an asymmetry between the way we assign income for non-farmers (which is solely based on the

1940 Census data) and the way we assign income for farmers (which combines information from

the 1940 and 1960 Censuses). Hence, we also test the robustness of our results to using the IPUMS

variable occscore, which corresponds to median earnings in a given occupation in the 1950

Census. While this measure uses less information (ignoring, for instance, any immigrant penalty

within the first generation), it is computed in a symmetric way for farmers and non-farmers. In

addition, we check the robustness of our results to alternative measures of income (for both farmers

and non-farmers) in Section 5. Our main qualitative conclusions are unaffected by the specific

proxy for income that we use for the historical samples, although we tend to find larger

intergenerational gaps when using the more complete set of characteristics to predict income.

17 The predicted earnings we estimate are very similar when we use the machine-learning approach (i.e., the lasso algorithm) to choose the most important predictors of earnings in the initial regression; the correlation between the predicted earnings and the lasso-estimated predicted earnings for fathers in the 1910–1940 cohort is 0.99. 18 CW use a ratio of farmers to farm laborers by region; we add an additional adjustment by immigration status. We further follow CW by scaling up the earnings of farm laborers and farm managers to account for non-cash (in kind) earnings. Throughout these calculations, we restrict the sample to US-born whites and immigrants from the most common sending countries during our historical time period. For more details, see Section 3 in the Online Appendix. 19 US-born fathers who are farmers are on average around the 40th and 36st percentile of the income distribution in 1880 and 1910, respectively. Immigrant fathers who are farmers are on average around the 38th and 32th percentile of the income distribution, respectively.

11

For sons observed in 1940, we conduct our analysis using two income measures. First, we use

our predicted income for these sons so as to use a consistent income measure across all historical

samples and Census years. As an alternative approach, we use individual income data from the

1940 Census. Note that this measure includes only wage and salary income, and does not include

capital income or income from self-employment (including farm income). We therefore also use

the CW approach for adjusting the income of farmers and all other self-employed individuals

described above. In this case, we calculate the ratio of mean earnings for self-employed non-wage

workers in an occupation to the mean earnings of wage workers in the same occupation from the

1960 Census and then use this ratio to scale wage earnings of the self-employed in the 1940

Census.

One limitation of using the 1940 Census to compute income scores for earlier Censuses is that

we rely on the relative rank between our income cells to be stable from 1880 to 1940. This is

especially a concern for our 1880–1910 cohort. To address this, we show the robustness of the

results to an alternative data source that measures earnings in an earlier period, namely the 1901

Cost of Living Survey. We also note that Saavedra and Twinam (2018) show that earnings ranks

across occupations are fairly stable for later cohorts covering the course of the 20th century.

B. Modern Datasets for Studying Economic Mobility

Our linked Census data cannot be extended beyond 1940 because confidentiality restrictions

dictate that individuals’ names are only released to the public 72 years after a Census is taken.

Instead, we rely on two sources of data to follow modern father-son pairs in the labor market. First,

we use publicly available aggregate data from the Opportunity Insights project, which was

originally constructed by linking US Census and American Community Survey data with federal

income tax returns and is based on large samples (Chetty et al. 2018a).20 Second, to address the

discrepancies between historical and modern data in the use of occupation versus income

measures, we use the General Social Survey (GSS), a smaller, survey-based dataset that asks

respondents about their own occupation and the occupations of their parents.

Our sample from the Opportunity Insights project is based on nearly 3 million male children

born between 1978 and 1983, out of which 160,000 have an immigrant father born in one the

12

twenty-one largest sending countries to the US.21 These data are built by the Opportunity Insights

team in two steps: In the first step, the 2000 US Census long form and the 2005-2015 American

Community Survey (ACS) are linked to federal income tax returns (using a unique person

identifier assigned by the Census Bureau staff). This step enables us to observe an individual’s

income from the tax records as well as his or her race and place of birth from the Census or the

ACS. In the second step, we obtain information on family income during childhood. To do so, an

individual’s parent is defined as the person who first claims him or her as a dependent in a federal

income tax return at some point between 1994 and 2015. Note that as a result of this procedure for

linking parents to children, we are not able to observe pairs for which either the child or the parents

lack a Social Security Number. As a result of this limitation, the sample is restricted to children

who are either US citizens or authorized immigrants who came to the US during childhood, and

whose parents are also US citizens or authorized immigrants.

Adult income for our childhood cohort is measured as the average of annual income in 2014

and 2015, when these children were between the ages of 31 and 37, whereas parental income is

measured as the average household income from 1994 to 2000 (fifteen to twenty years before the

children are observed in the labor market). We used the publicly available version of the data,

which reports information collapsed by ventile bin in the national income distribution. Hence, for

each parental ranking, we observe the average income rank of children whose parents fall into that

ventile, by a father’s country of origin.22 We also show results when we instead use information

on children of foreign-born and US-born mothers. Using mothers instead of fathers allows us to

show results separately by race and by percentile rather than ventile bins (but in that case the data

do not allow us to show results by country of origin).23

These data have several attractive features for the study of the intergenerational mobility of

immigrants: It is very large relative to other panel datasets (no other longitudinal dataset would

21 This sample is smaller than the baseline sample in Chetty et al. (2018a) because the question on parental place of birth is only included in the ACS 2005–2015 and the Census 2000 long form (which covers a random sample of approximately one-sixth of US households). For this modern cohort, the 21 source countries are Canada, China, Colombia, Cuba, the Dominican Republic, Ecuador, El Salvador, Germany, Greece, Guatemala, Haiti, India, Israel, Italy, Jamaica, Japan, Mexico, the Philippines, South Korea, the United Kingdom, and Vietnam. 22 These data correspond to the “Non-Parametric Estimates of Income Ranks for Second Generation Immigrant Children by Parent Income, Country of Origin, and Gender” file available on the Opportunity Insights website at https://opportunityinsights.org/data/. 23 Specifically, we use the data in the “National Statistics by Parent Income Percentile, Gender, and Race” file available on the Opportunity Insights website at https://opportunityinsights.org/data/.

13

enable us to look at mobility differences by country of origin in the modern period),24 and includes

administrative data on income in both the parents’ and the children’s generation.25 However, the

dataset has two limitations for our purposes. First, it does not include information on individual-

level occupations—the main labor market outcome we observe in the historical data—thus

complicating our past-present comparisons of intergenerational mobility. Second, because

unauthorized immigrants do not have a Social Security Number (SSN), these data exclude the

children of unauthorized immigrants, a sizable fraction of the immigrants in the modern period.

This limitation affects Hispanic immigrants more than other groups. Specifically, Chetty et al.

(2018a) compares these data to the American Community Survey and reports that it includes about

79% of Hispanics, whereas the coverage is close to 100% for other ethnic groups. We conduct

subsample analysis by ethnicity and race (Hispanic, Asian and White), as well as by country of

origin, and concerns about coverage of the undocumented would not affect results for many of

these country-of-origin groups. Furthermore, we add linked data from the General Social Surveys

and cross-sectional data from the Current Population Survey (both of which include unauthorized

immigrants) to partially address this concern.

As we mentioned, we supplement the Opportunity Insights data with survey data from the

General Social Survey. Although the GSS has substantially smaller sample sizes, it includes

questions on occupation and covers unauthorized immigrants, both of which facilitate comparison

with the historical data. The survey includes information on a respondent’s occupation, as well as

parents’ place of birth (in the US or abroad) and retrospective questions on parents’ occupations

when the respondent was 16 years old. To make this sample as comparable as possible to the

historical data, we focus on respondents who: (1) are males, (2) lived in the US by age 16, (3) were

30 to 50 years old at the time of the survey (which implies an average gap of approximately 24

years between parents’ and sons’ occupations), and (4) have available information on his and his

father’s occupation. We attempt to maximize comparability with the Opportunity Insights sample

while retaining enough observations by restricting the analysis to survey years 2000 to 2018. After

24 For instance, the PSID (which includes information on parental income) started in 1968 and thus does not have much coverage of immigrant families who arrived to the US after the border reopened to immigration in 1965. The sample has been updated in recent years and now includes about 500 immigrant families, although the survey documentation warns that “it is important to note that the size of the immigrant refresher sample was driven by budget constraints and is not designed for immigrant-focused subgroup analysis.” 25 Other panel datasets (for instance, the National Longitudinal Surveys of Youth used in Borjas 1993) include retrospective questions (asked of the children) about the occupation and education in the parent’s generation, but do not contain information on parental income.

14

imposing these sample restrictions, our GSS sample includes about 3,100 observations, out of

which about 9% are second-generation immigrants.

To assign income proxies in the GSS data, we implemented an analogous procedure to the

creation of the historical proxy, with minor adjustments to fit the available variables.26

Specifically, we used the 2006–2015 Current Population Surveys and regressed logged total earned

income on a number of fixed effects and complete set of interaction terms using occupation, age,

region, and immigrant status (and their interaction terms) as our explanatory variables. We then

used this regression to predict income scores for respondents and their fathers.

3. Preliminary Evidence: Earning Gaps between Immigrants and US-born since 1880

We start by estimating earning gaps between US-born men and first- and second-generation

immigrants by country of origin. For this exercise, we use cross-sectional samples, rather than

father-son pairs. Our interest is in documenting the earnings disadvantage faced by first-generation

immigrants and to ask whether the children of immigrants are able to erase some or all of these

gaps by the next generation.

We estimate the following equation, first for the father’s generation and then for the son’s:

(1) 𝑌𝑌𝑖𝑖𝑖𝑖𝑖𝑖 = 𝛼𝛼 + ∑ 𝛽𝛽𝑖𝑖𝑁𝑁𝑖𝑖=1 Countryc + 𝛿𝛿𝑖𝑖 + 𝜖𝜖𝑖𝑖𝑖𝑖𝑖𝑖.

where 𝑌𝑌𝑖𝑖𝑖𝑖𝑖𝑖 is the logged income score of father or son 𝑖𝑖 of age 𝑎𝑎 from sending country 𝑐𝑐.27 The 𝛽𝛽𝑖𝑖

coefficients on the country-of-origin fixed effects reveal the average income-score difference

between US-born individuals and immigrants from country 𝑐𝑐. All regressions include a quadratic

in father’s or son’s age 𝛿𝛿𝑖𝑖.

Figure 1 plots the 𝛽𝛽𝑖𝑖 coefficients for first-generation (in black) and second-generation (in gray)

immigrants in each cohort.28 For most countries-of-origin and time periods, the gaps in income

26 The GSS reports parental immigrant status but does not record country of origin. The most precise measure of geography in the GSS is region. In addition, the GSS does not include information on parental age when the respondent was 16 years old (when parental occupations are measured) or at any other time. 27 We note that the results are very similar when use actual (rather than predicted) income for the 1940 sons as well as for individuals in the modern cohort (See Appendix Figure A1). 28 In particular, we use the 1980 Census for the fathers’ generation and the Current Population Survey (CPS)’s Annual Social and Economic Supplement for 2006–2015 for the sons (Flood et al. 2018, Ruggles et al. 2018). We restrict the 1980 Census sample (fathers) to men who are aged 30–50, who have a son present in the household, and who were born in the US or in one of the top sending countries identified in the Opportunity Insights sample. We restrict the CPS sample (sons) to US-born men who are aged 30–50 and whose fathers were either born in the US or in one of the top sending countries. For both of these datasets, we construct occupation-based income scores using an analogous procedure to the one that we use in both the GSS and the historical data, but we also show the results from using actual income.

15

scores between first-generation immigrants and US-born men range between -20 log points and

+20 log points. In the modern cohorts, four countries faced larger gaps of nearly -60 log points

(the Dominican Republic Haiti, Mexico, and Vietnam). In the past, immigrants from Scandinavian

countries faced the largest earnings penalties, along with immigrants from Russia and Italy (1880)

and from Portugal and Italy (1910). In all cohorts, immigrants from a set of sending countries

outperform the US-born in the labor market. Today, these high-earning countries include

Germany, Canada, the United Kingdom, Japan, and India, whereas in the past the highest-earning

immigrants were from the England, Scotland, Germany, and France.

In each of the three cohorts, second-generation immigrants from countries in which first-

generation immigrants earned less than the US-born close or even reverse the corresponding

earnings gap in the second generation (so that children of immigrants begin to out-earn children

of US-born individuals). There are only a few cases in which the gap remained above 5 log points

by the second generation (Norwegians in 1910 and 1940, Finish in 1940, and Jamaicans, Haitians,

and Mexicans in the modern data), but, for most sending countries, the earnings gap is close to

zero by the second generation. Appendix Figure A2 shows the same pattern when we consider the

daughters of immigrants. In this case, because historically many women were not in the labor

force, we assign to women the sum of their own incomes (if she was employed) and the incomes

of their husbands (if she was married); the earnings gaps are, again, either closed or reversed by

the second generation in each of the three cohorts.

For countries whose first-generation immigrants already out-earned US-born individuals in the

labor market, the pattern is more mixed, with the corresponding earnings advantage becoming

smaller for some countries, remaining similar in size for others, or even growing by the second

generation in some cases. Although earnings gaps for countries that start below the US-born

diminish substantially across generations, it is still the case that the children of low-earning

immigrants tend to earn less than the children of high-earning immigrants, consistent with the

persistence results in Abramitzky, Boustan, and Eriksson (2014) and Ward (Forthcoming).29

29 The cross-sectional samples in Figure 1 have three main advantages: they are constructed to be nationally representative, they do not rely on linking and they are highly comparable over time. However, the drawback of cross-sectional analysis is that it does not allow for a direct comparison of fathers with their sons. In addition, cross-sectional data will include first-generation immigrants who might have returned to their home country before having children or with their families (so that their sons would not be in the subsequent sample thirty years later). Because return migrants tend to be negatively selected (Abramitzky, Boustan, and Eriksson 2014, 2018; Lubotsky 2007), this figure might over-state the degree of intergenerational improvement by the second generation. Appendix Figure A3 shows that the same pattern holds in our linked father-son samples, namely the sons of immigrants are on average more

16

4. Main Results: Intergenerational Mobility of Immigrants since 1880

We have so far considered average earnings gaps, without any regard for the parent’s rank in

the income distribution. The earnings growth for the children of immigrants relative to the children

of US-born parents in Figure 1 could be driven by higher rates of upward mobility from the bottom

of the income distribution, lower rates of downward mobility from the top of the income

distribution, or both. The essence of the American Dream is that even immigrants at the bottom of

the income distribution can have children that succeed. Thus, we turn to the linked data, both

historical and modern, to study intergenerational mobility of immigrants and US-born individuals

whose parents had comparable labor market outcomes as measured by their rank in the national

income distribution (following Chetty et al. 2014 and Chetty et al. 2018a). We first rank each son

based on his income score (or actual income when using the Opportunity Insights data), relative

to other sons born in the same year. Similarly, we rank fathers based on their income score relative

to all other fathers with sons born in the same year.30

In particular, we regress a son’s rank on his father’s rank31, allowing both the slope and

intercept to differ for sons of immigrants and US-born individuals:

(2) 𝑅𝑅𝑎𝑎𝑅𝑅𝑅𝑅 𝑆𝑆𝑆𝑆𝑅𝑅 = 𝛼𝛼 + 𝛽𝛽02𝑅𝑅𝑛𝑛 𝐺𝐺𝐺𝐺𝑅𝑅. + 𝛽𝛽1𝑅𝑅𝑎𝑎𝑅𝑅𝑅𝑅 𝐹𝐹𝑎𝑎𝐹𝐹ℎ𝐺𝐺𝑒𝑒 + 𝛽𝛽22𝑅𝑅𝑛𝑛 𝐺𝐺𝐺𝐺𝑅𝑅.× 𝑅𝑅𝑎𝑎𝑅𝑅𝑅𝑅 𝐹𝐹𝑎𝑎𝐹𝐹ℎ𝐺𝐺𝑒𝑒 + 𝜖𝜖.

In the terminology of Chetty et al. (2014), the constant term 𝛼𝛼 measures absolute rank mobility for

the children of the US-born, that is, the expected rank of children of US-born individuals with

fathers at the very bottom of the income distribution. The parameter 𝛽𝛽0 captures the degree to

which this expected rank is different for the children of immigrants. The parameter 𝛽𝛽1 measures

the rate of relative mobility of children of US-born individuals, or the association between the

upwardly mobile than the sons of US-born individuals. We also re-run this specification but excluding individuals living in the South in either generation in order to estimate more-comparable earnings gaps (since 95 percent of foreign-born individuals lived outside of the South). Appendix Figure A4 displays the results, showing that the main findings of this section are preserved. 30 If there are individuals in the sample who have the same father, we assign the father the average rank across his sons. Due to the GSS’s much more limited sample size compared to the historical linked data and the Opportunity Insights data, we pool all survey years (2000 to 2018) and include a survey year fixed effect in the regressions. 31 In the Opportunity Insights data, we use family income rank because we do not have separate information on a father’s income rank.

17

ranks of children and those of their fathers. Finally, 𝛽𝛽2 measures the degree to which this

association is different for children of immigrants.32

We compare the economic mobility of the sons of immigrants to the sons of the US-born in

Figure 2, which plots these regression lines estimated in equation (2), as well as a binned scatterplot

showing the mean income rank of the children of immigrants and of the US-born by parental

income rank. Figure 2 contains four panels, two for the historical cohorts and two for the modern

cohort using the GSS and Opportunity Insights data, respectively. In each case, second-generation

immigrants were more upwardly mobile than the children of the US-born: conditional on the

father’s rank (or family’s rank when using the Opportunity Insights data), children of immigrants

achieve a higher expected rank than the children of the US-born. The mobility patterns of the

historical cohorts, shown in panels (a) and (b), suggest that, in the past, the sons of immigrants on

average out-earned comparable sons of US-born individuals throughout the income distribution,

and particularly in its bottom half. Second-generation immigrant children with fathers at the 25th

percentile in 1880 and 1910 reached an average income rank that was 3–4 percentiles above the

average rank reached by children of the US-born with similar fathers’ incomes.33

Panels (c) and (d) of Figure 2 show the mobility patterns for the modern cohort when using the

GSS and Opportunity Insights data, respectively. Today, the children of immigrants remain more

upwardly mobile than their counterparts with US-born parents, but this pattern mainly holds in the

bottom half of the income distribution. The children of the US-born in the 25th percentile rank at

the 4th percentile.34 Similar to the historical data, sons of immigrants in the 25th percentile rank

about 5 percentile points higher than children of US-born individuals, regardless of whether we

use the GSS or the Opportunity Insights data.35 In Table 1, we report the estimated values of the

32 A negative 𝛽𝛽0 implies a lower expected rank for children of immigrants relative to children of US-born individuals with parents at the very bottom of the income distribution. Note, however, that even if 𝛽𝛽0 was negative, the cross-sectional data could show convergence by the second generation for groups that started below the US-born (due to mean reversion). By a similar logic, immigrant groups that started above the US-born might lose some of their advantage by the second generation even if 𝛽𝛽0 was positive. 33 Appendix Figure A5 shows that the mobility advantage of the children of immigrants continues into the next generation. That is, the children of US-born individuals with foreign-born grandfathers are more mobile than the those with US-born grandfathers. 34 Our calculation is very similar but not identical to Chetty et al. (2018a, Figure 3), since we define immigrant status using the father’s place of birth and they define it using the mother’s. 35 Appendix Figure A6 shows the same rank-rank results but using actual (instead of predicted) income when it is available (i.e., for the 1940 sons and for the sons in the GSS cohort). This figure shows that replacing predicted with actual income does not alter the main findings for the 1910–1940 and GSS cohorts, and therefore suggests that our finding of higher upward mobility of children of immigrants is not driven by the use of predicted rather than actual income.

18

intercepts and slopes for each of the samples. For all three cohorts, we find a higher intercept for

the children of immigrants relative to the children of the US-born, indicating higher levels of

absolute mobility (these differences are statistically significant). The difference in intercept range

from 4 percentiles in the 1910–1940 cohort to 7 percentiles in the modern Opportunity Insights

data. The slopes for the children of immigrants range from 0.21 to 0.37 and are in most cases

smaller in magnitude than the slopes for US-born individuals, also suggesting a weaker association

between a father’s rank and his son’s rank for immigrant families.

In Figure 3, we allow the intercept and the slope of the rank-rank relationship to differ by

country of origin, both in the historical and the modern data. We then use these estimates to

compute the expected income rank for children born at the 25th percentile for each country of

origin.36 The figure shows that children from nearly all sending countries in the past and the present

(except Norway and Belgium in 1880, Norway and Germany in 1910, and Haiti and Jamaica in

the modern cohort) had a higher expected rank than the children of US-born individuals with

parents at the 25th percentile of the income distribution.37

Figure 4 shows that the children of immigrants and the US-born appear more similar when

looking at the 75th percentile of the income distribution.38 In the past, children of immigrants had

slightly higher expected ranks, and today the two groups achieve the same expected rank. Children

of immigrants from some sending countries like Portugal and Belgium in the past and Jamaica, El

Salvador and Mexico today did worse than the children of US-born, and those from other countries

like Russia and Ireland in the past and India and China today did better.

We next consider differences in mobility patterns by immigrant ethnicity or race (white,

Hispanic, and Asian) within the modern cohort, again using the Opportunity Insights data.

Consistent with our country-by-country results, Figure 5 shows that immigrants from all of these

groups are more upwardly mobile than US-born individuals (where the data for US-born

individuals pools all races), particularly in the bottom half of the income distribution.39 White and

36 In this exercise, we focus on the Opportunity Insights for the modern data because of its large sample size, which facilitates an analysis of sub-samples, and because the GSS data does not record parent’s country of origin or ethnicity. 37 Appendix Figure A7 shows the same figure, but weighting observations in the historical cohorts to account for selection into the linked sample using the individual’s childhood characteristics (like father’s birthplace and occupation, state of residence, etc.). Appendix Table A4 shows the intercept and slope for each country of origin in the two historical cohorts. 38 The modern data do not include information on the children of families at the 50th percentile by country of origin. 39 We note that in this case we cannot restrict the sample to the 21 largest sending countries since the data is only disaggregated by race and not by country of origin. We also note that in this case second-generation status is defined based on the birthplace of the mother.

19

Hispanic immigrants perform similarly to US-born individuals in the top half of the distribution

(panels (a) and (b)). By contrast, second-generation Asian immigrants can expect to reach higher

ranks than US-born individuals, even at the top of the income distribution.40

Overall, we conclude that second-generation immigrants are on average more upwardly mobile

than the sons of US-born individuals. This advantage is not driven by immigrants from countries

with higher mean earning than the US-born in the first generation: there is higher upward mobility

even among immigrants from countries that start significantly below the US-born in the first

generation. In both the past and present, sons of immigrants move out of the bottom of the income

distribution at higher rates than the sons of US-born parents, with differences in absolute mobility

ranging between 3 and 6 percentiles. The similarities are striking especially given that in the past,

immigrants hailed from more-comparable economies (i.e., European countries) and today they hail

from poorer regions (i.e., Latin America and Asia).41

Section 4 of the Online Appendix documents mobility patterns in more detail using

occupational transition matrices. Overall, the analysis suggests that the children of immigrants

were more likely than the children of the US born to use urban routes to upward mobility. The

sons of unskilled immigrant fathers moved into white-collar and skilled blue-collar jobs at higher

rates than the sons of unskilled US-born fathers, who were more likely to transition into farming.

In the modern period, children of immigrants whose fathers were employed as blue-collar or

service workers were more likely to become professionals than the children of the US-born.

5. Why are children of immigrants more upwardly mobile?

This section explores why the children of immigrants were more upwardly mobile than the

children of the US-born. Whenever possible, we compare the historical and modern cohorts.

However, we often focus on the historical data for which we have a set of rich covariates because,

in the modern case, we only have access to aggregate cells in the Opportunity Insights data and a

limited sample size in the GSS. We consider two potential explanations: first, a “mismatch”

between an immigrant father’s full potential and his earnings rank—due, for example, to a lack of

40 This result echoes Chetty et al. (2018a)’s finding that exceptional outcomes of Asian children are driven by immigrants (i.e., individuals whose parents were not born in the United States). 41 US GDP per capita was more than 7 times higher than Mexico or China when migration flows from these countries began in 1970, whereas the US had GDP per capita that was only 2–3 times higher than European sending countries circa 1900.

20

US-specific skills. Such “mismatch” allows immigrant children more scope for upward mobility.

Second, differences in parental investments, including education and location choices.

A. Father’s income not reflecting earnings potential

One possible explanation for the higher upward mobility of immigrants is that the earnings

rank of an immigrant father in the US income distribution might not fully reflect his ability or

earnings potential. In other words, if immigrant fathers are ranked lower in the distribution than

their skills or ability would suggest (e.g., due to a lack of US-specific human capital, language

ability or labor market networks), then the children of these immigrants are expected to perform

better than their fathers’ outcomes would predict. We consider various aspects of the data to see

whether this underperformance in the US labor market might play a role in explaining immigrants’

higher levels of upward mobility.

i. Parental Age of Arrival in the United States

We first consider the age at which immigrant fathers arrived in the United States. Immigrant

fathers who arrived as children to the US would likely assimilate into the labor market more easily

(for instance, by being exposed to US education since an early age), so that their observed income

would be closer to their potential earnings. Moreover, an earlier age of arrival to the US might

benefit immigrants coming from less developed countries if they are exposed to better schools,

health facilities, etc. in the US (which would then increase their earnings ability as adults). Indeed,

Ward and Alexander (2018) shows that age of arrival was a key driver of immigrant earnings

during the Age of Mass Migration, with immigrants arriving at younger ages out-earning those

arriving later in life. If lack of US-specific human capital is driving our results, then we should

observe higher levels of upward mobility for children of immigrants who arrived in the US as

adults and thus may have been placed too low in the income distribution relative to their underlying

ability.

Figure 6 calculates the intergenerational gap at the 25th percentile of the income distribution

for three groups of immigrants: those whose fathers arrived to the US before age 7, those whose

fathers arrived between the ages of 8 and 16, those whose fathers arrived at age 17 or older. We

focus this analysis on the 1910–1940 cohort because the 1880 Census did not include information

on year of arrival to the US. Parental age of arrival to the US is also not available in the GSS. We

find that the intergenerational mobility gap between the children of immigrants and the US-born

21

grows with fathers’ age of arrival, so that the gap is more than twice as large for the children of

immigrants who arrived as adults than for the children of immigrants who arrived as children.42

The children of immigrants who arrived before schooling age and thus likely received all of their

schooling in the US exhibit no mobility gap relative to the children of the US-born. This result

suggests that part of the difference in upward mobility is driven by immigrant fathers not earning

up to their potential in the labor market.

ii. Parental English Ability

One important aspect of US-specific human capital is English language ability. If immigrant

fathers end up lower in the income distribution than their potential would predict due to an

imperfect command of English, then we should expect to see high levels of upward mobility for

their children, most of whom would have been educated in schools that used English as the primary

language of instruction.

To test this hypothesis, we compare the association between upward mobility and father’s age

of arrival for fathers from English- versus non-English-speaking countries; this method is similar

in spirit to Bleakley and Chin (2010). All immigrants will have greater exposure to American

culture and networks if they arrive in the country at a younger age. Immigrants from non-English

speaking countries will additionally benefit from a greater command of the English language. If

English language ability can partially explain fathers’ mismatch in the income distribution, we

should find a stronger relationship between father’s age of arrival and upward mobility for the

children of immigrants from non-English speaking countries. Indeed, Figure 7 shows that, while

the intergenerational mobility gap grows with fathers’ age of arrival for both groups, the age

gradient is steeper for individuals with fathers from non-English-speaking countries. This result

suggests that lack of English ability partially contributes to the misplacement of immigrant fathers

in the income distribution. We note, however, that individuals with English-speaking immigrant

fathers still have significant intergenerational gaps (of at least 5 percentage points), implying that

lack of English ability cannot fully explain immigrants’ higher levels of upward mobility in the

past.

B. Did immigrants invest more in the upward mobility of their children?

42 We also include the corresponding rank-rank figure (i.e., Appendix Figure A8).

22

i. Investment in Education

We first study the possibility that children of immigrants achieved higher levels of educational

attainment, which then served as a vehicle toward achieving better labor market outcomes in

adulthood. Figure 8 considers the 1910–1940 cohort—which is the first cohort for which we have

data on completed years of schooling for the sons—and plots the average educational attainment

of the sons of immigrants and of US-born individuals relative to their fathers’ income rank. Panels

(a) and (b) of this figure show that, if anything, sons of US-born fathers were more likely to

graduate high school and reach higher grades in school. In the same vein, panel (c) focuses on

linked sons aged 12–16 in 1910 to show that the sons of US-born fathers were indeed more likely

to be attending school in 1910 than the sons of immigrant fathers.43 Hence, higher educational

investments among immigrant families are unlikely to be the explanation for their higher upward

mobility.

The fact that educational differences do not explain the intergenerational mobility gap is not

entirely surprising because, as Goldin (1998) shows, the returns to schooling in 1940 were lower

than in recent years. The children of immigrants also enjoyed higher income mobility because they

earned a higher income at any given level of education (panel (d)). However, we also note that the

children of immigrants enjoyed a faster rate of educational mobility, even if they did not achieve

higher levels of educational attainment. That is, for a given level of father’s education, the children

of immigrants in the bottom half of the distribution tended to achieve more years of schooling than

the children of the US-born, consistent with the finding by Card, Domnisoru, and Taylor 2018 (see

panel (e)).44

ii. Location Choices

We next consider the role of location choices in explaining immigrants’ higher levels of

upward mobility. Our analysis of location choices is motivated by two facts. First, rates of

intergenerational mobility both historically and today vary greatly by region (Chetty et al. 2017,

Tan 2018, Connor 2019). Second, previous research shows that first-generation immigrants are

43 Similarly, using the 10% sample of the 1880 Census, we see that sons of immigrant fathers who are aged 12–16 in the earlier cohort are 8 percentage points less likely to be attending school than sons of US-born fathers (57 versus 65 percent). 44 Panel (e) is based on father-son pairs for whom we could link the father to the 1940 Census and thus have a measure of father’s education (22 percent of the sample).

23

more responsive than the US-born to wage differentials across regions (Borjas 2001, Cadena and

Kovak 2016), which raises the possibility that they might also be more likely to move to areas with

better prospects for their children.

As motivation, Figure 9 presents maps that indicate the share of children who are second-

generation immigrants in 1880 and 1910 by State Economic Areas (SEAs). Immigrant families

typically settled in the Northeast, the upper Midwest, and some parts of the West, and were very

unlikely to settle in the South (see panels (a) and (c)). Panels (b) and (d) then depict the share of

each SEA’s population that was upwardly mobile, focusing on the sons of US-born fathers (to

avoid a mechanical relationship between upward mobility and the fraction of immigrants).45 Our

measure of upward mobility is similar in spirit to that of Chetty et al. (2018b): we calculate the

share of sons who reach the top third of the income distribution, conditional on having had a father

who was in the bottom third of the corresponding national distribution. The maps show that the

areas in which immigrants settled tended to have overall higher levels of upward economic

mobility.

The final two panels of Figure 9 indicate the share of the population who is foreign-born in

1980 by county (using data from Manson et al. (2019)), and the share of each county’s white non-

Hispanic male population that was upwardly mobile using data from Chetty et al. (2018b).46

Immigrants settled in the Northeast and the West, as well as in Florida, but were less likely to live

in the Midwest or the South. As for earlier cohorts, immigrants choose to settle in areas with high

levels of upward economic mobility.

We summarize the information in these maps in Figure 10, in which we plot the share of each

SEA’s population that is a second-generation immigrant, against the share of sons of the US-born

raised in that SEA who achieved upward mobility. This figure confirms that first-generation

immigrant parents were more likely to settle in areas with higher mobility prospects, both in the

historical and in the modern data.47 These figures therefore suggest that immigrants might exhibit

45 Appendix Figure A9 shows the share of each SEA’s population that was upwardly mobile, separately for the sons of US-born and foreign-born fathers. 46 We are not able to compute the fraction of upwardly mobile sons separately for the children of US-born and foreign-born parents at the commuting zone level in the modern data. Instead, we used publicly available data indicating the fraction of non-Hispanic white males who reach the top 20 percent of the income distribution, conditional on having parent income in the 25th percentile of the income distribution (Chetty et al. (2018b)). 47 The first two panels of Appendix Figure A10 reproduce these figures at the state level for the historical cohorts. The third panel reproduces this figure at the commuting zone level for the modern cohort. All of these graphs reinforce the positive relationship between upward mobility and immigrant location choices. Appendix Figure A11 plots the share of sons of the US-born who are upwardly mobile in a state or SEA against the corresponding share of second-

24

higher levels of intergenerational mobility because they tended to migrate to areas that had overall

higher degrees of upward mobility.

We now explicitly test the hypothesis that differences in the geographic distribution of

immigrants and US-born individuals might explain mobility gaps between the two groups.

Specifically, we re-run our rank-based specification from equation (2), and sequentially add

region, state, state-by-urban status, and county fixed effects (based on an individual’s childhood

location). These specifications allow us to compare immigrants and US-born individuals with

similar parental incomes and who grew up in the same location.

Figure 11 plots the baseline intergenerational gap between the children of immigrants and the

US-born at the 25th percentile for each cohort and how the gap changes as we incorporate location

controls. The second bar of panels (a) and (b) shows that including childhood region fixed effects

diminishes the intergenerational gaps by roughly 70% in both historical cohorts. The

intergenerational gaps are further reduced when adding childhood state fixed effects (third bar of

each graph). Finally, when we further include either childhood state-by-urban status fixed effects

(fourth bar) or childhood county fixed effects (fifth bar) the gap between the children of

immigrants and the US-born fully closes. Hence, these results suggest that location choices are an

important source of difference in upward mobility for the children of immigrants versus children

of the US-born.48

The GSS data enable us to explore the role of location choices in the modern period. While the

survey does not include detailed childhood occupation as in the historical data, it includes a

question on a respondent’s region of residence at age 16. Panel (c) of Figure 11 shows that the

intergenerational gap is unchanged once comparing children growing up in similar locations.

Relative to the historical data, the choice of region appears to be less important in this cohort, but

generation immigrant sons in that same state or SEA who achieved upward mobility in the historical cohorts. Because many SEAs have small populations in the linked sample, the SEA estimates in these plots are much less precise. Nevertheless, we notice that the same places that are typically associated with upward mobility for immigrants seem to have comparable levels of mobility for US-born individuals (as evidenced by the clustering of the estimates around the 45-degree line) and that the Northeast and West are areas with higher levels of upward mobility. 48 Appendix Figure A12 plots the intergenerational gap between the children of immigrants and the US-born at the 50th and 75th percentile for each cohort and how the gap changes as we incorporate location controls. These results highlight the importance of first-generation immigrants’ location choices throughout the income distribution. Our baseline income scores use state of residence to predict income, so we were concerned that adding state fixed effects will mechanically have a strong predictive power. However, Appendix Figure A13 shows that the results are similar when using income scores that do not use state to predict incomes.

25

it is possible that county or neighborhood choice (which we cannot capture) are still important

today.

We also consider the potential role of enclaves in boosting or hindering mobility of the second

generation. To do so, we classify a second-generation immigrant as growing up in an enclave if at

least 10 percent of the adult population in his childhood enumeration district came from the same

country as his father (enumeration districts are the smallest level of Census geography in this

period and are roughly the same size as Census tracts today). We then calculate the

intergenerational gap separately for those who do and do not grow up in immigrant enclaves.

Appendix Figure A14 shows that in the historical cohorts, immigrant children who did not grow

up in an enclave had higher rates of upward mobility than children growing up in enclaves.

Because living in an enclave is likely correlated with other characteristics that might matter for

mobility, we emphasize that this evidence is only suggestive of the potential role of enclaves.

A useful comparison group to shed light on the relative roles of geographic choices and human

capital is the children of US-born fathers who migrated internally within the US. On the one hand,

these families were less likely to suffer an “immigrant” penalty in the first generation (due to

discrimination or a lack of US-specific human capital). On the other hand, they were likely to sort

into areas with better prospects for their children. Indeed, Appendix Figure A15 shows that the

upward mobility of children of US-born fathers who migrated internally falls between children of

immigrants and children of the US-born fathers who remained in their state of birth, suggesting an

important role for geographic choice.

Finally, we consider the location choices of the second generation to explore whether the sons

of immigrants are themselves more likely to move as adults. Appendix Figure A16 shows that sons

with immigrant fathers throughout the income distribution were less likely to move away from

their childhood county than sons with US-born fathers. This pattern is consistent with the idea that

second-generation immigrants were already living in areas with relatively better prospects, and

thus might have felt less compelled to move as adults.

6. Robustness

This section assesses the robustness of our main rank-rank mobility estimates to the choice of

Census linking procedure, the method of assigning income by occupation and other attributes, and

the details of sample construction.

A. Sensitivity of Results to Linking Procedure

26

Our baseline Census linking strategy uses a standard algorithm based on Abramitzky, Boustan

and Eriksson (2012, 2014), which standardizes first and last names into phonetic equivalents to

account for transcription errors and allows ages to differ by up to two years to account for age

misreporting. We consider the robustness of our main finding to alternatives for processing name

and age data, and to choice of algorithm itself. Specifically, we re-run the rank-based specifications

using three matching strategies that are more conservative than our baseline strategy. First, we

restrict the sample to exact matches by full (non-standardized) name and age. Second, we restrict

the sample to individuals whose first and last names are unique within a five-year age band. Third,

we require matches to be both exact and unique within a five-year age band. Because each of these

linking procedures is more conservative than our baseline one, the samples from these approaches

will be smaller, but likely contain fewer matching errors.

Appendix Figures A17 and A18 show the rank-based results using each of the three more-

restrictive matching strategies. Regardless of the linking procedure we use, the main result that

immigrants are more upwardly mobile than US-born individuals throughout the income

distribution remains unchanged. Moreover, the magnitude of the intergenerational gaps is also very

close to the magnitude in the baseline sample for both cohorts, at roughly 3–4 percentile points for

children with parents in the 25th percentile. Appendix Table A5 shows the estimated values of the

intercept and slope coefficients in these more conservatively linked samples for both cohorts. In

all cases, the values are close to those that we obtain in the baseline sample and confirm our

conclusion of higher upward mobility among children of immigrants.

Finally, we consider the possibility that our result could be explained by selection into the

linked sample. Specifically, we re-estimate the rank-rank specification after reweighting the

matched data by characteristics such as age, state of residence, birthplace, and occupation, thereby

accounting for selection into the matched sample, at least with respect to observable

characteristics.49 Appendix Figure A19 shows that the results are very similar when using the

reweighted data.

49 In this exercise, weights represent the inverse probability that an individual can be linked across Censuses based on his observable characteristics. For each individual, we can construct the weights using either observable characteristics from childhood (which include his father’s characteristics, like birthplace and occupation) or from adulthood (which include his own occupation). We cannot combine the two sets of characteristics into a single weighting function because we only know both sets of characteristics for individuals who are successfully linked. For more details on the construction of these weights, see Section 3 in the Online Appendix.

27

Overall, these exercises indicate that our main finding – namely, that children of immigrants

are more upwardly mobile than children of US-born individuals – is robust to various features of

the linking procedure.

B. Sensitivity of Results to Income Assignment

In this section, we check the robustness of our historical results to various proxies of income.

First, we produce alternative measures of income for farmers, a large portion of the economy for

which historical income data is less accurate. Second, we drop fathers and sons who were farmers

from the sample. Third, we replace our income proxy, which is based on the 1940 Census, with

information collected closer to the years in which the fathers in our sample would have been in the

labor market. Finally, we create different income predictions from the 1940 Census that are based

on more and less information about individuals. These results are summarized in Appendix Tables

A6 and A7, which show—for our baseline measure and for our various robustness measures—the

slope and intercept from the rank-rank specification for the children of US-born individuals and

the children of immigrants from each country.

We first consider how our estimates change when we use an alternative measure of farmer’s

income using the 1900 Census of Agriculture. In particular, we assign all farmers a measure of

income based on revenues and expenditures from the 1900 Census of Agriculture at the county

level to account for the wide variation in crops and farm size by location.50 This measure does not

allow farmers’ income to differ by immigration status, which may overstate the income of first-

generation immigrant farmers and thus understate the mobility of their sons. The Census of

Agriculture adjustment, on average, places farmer fathers in the 48th percentile of the income

distribution in 1880 and 1910, which is higher than the average ranks using our preferred Collins

and Wanamaker (CW) method. The first panel of Appendix Figures A20 and A21 show the rank-

rank specifications using this adjustment for farmer income. Similar to our results above, the figure

indicates that in both cohorts, immigrants can on average expect to reach a higher rank in the

earnings distribution than US-born individuals growing up with comparable incomes.

50 Specifically, we use the 1900 Census of Agriculture to calculate net earnings for farmers at the county level and adjust these calculations to 1940 dollars. This calculation follows a similar approach to that in Goldenweiser (1916) and Mitchell et al. (1922), and then later used in Abramitzky, Boustan, and Eriksson (2012). For more details, see Section 3 in the Online Appendix.

28

Second, given the prominence of farming and the difficulty in assigning farmers income, we

consider how our main results change when we drop fathers and sons who are farmers from the

sample. Panel (b) of both figures display the results from re-running equation (2) with ranks from

the non-farming national income distribution. We see that the higher degree of upward mobility

that we find for sons of immigrants, especially in the bottom of the income distribution, is not

driven by individuals in farming occupations alone. Of course, moving into or out of farming was

by itself an avenue for mobility, which is why we prefer to include farmers in our baseline sample

despite the challenges in measuring farm income.

Third, for individuals in the 1880 and 1910 Censuses, we replace our baseline income measure,

which is based on the 1940 Census, with information from the 1901 Cost of Living Survey (Preston

and Haines 1991).51 Unlike our baseline measure—which varies by occupation, region, state, age,

and country of origin—the income measure from the 1901 survey varies only by occupation and

was collected exclusively in urban areas. The benefit of this survey is that it was conducted closer

to the years in which our sample was in the labor market (1880 and 1910). Panels (c) of Appendix

Figure A20 and A21, which combine data from the 1901 Cost of Living Survey for non-

agricultural occupations with the 1900 Census of Agriculture measures of farm income, also show

that the children of immigrants remain more upwardly mobile than the children of US-born

individuals.

Fourth, we consider the sensitivity of the results to the variables included in the income

prediction in the 1940 Census. In our preferred approach, we use the relevant individual-level

information available in the Census to predict a person’s income. We next show the sensitivity of

the results to progressively using less information in the prediction. Panels (d) of the same figures

re-run the specifications using a measure of income for the fathers that does not incorporate an

immigrant penalty. Analogously, panel (e) of the same figures incorporates the father’s country of

origin into the son’s income prediction. We see that in both cases, sons of immigrant fathers

continue to be more upwardly mobile than comparable children of US-born fathers, with the

exception of panel (e) in the 1880–1910 cohort where there is no immigrant advantage at the very

bottom.

51 The 1901 Cost of Living Survey is closer in time to the earlier Censuses but less detailed than our baseline measure (based on the 1940 Census): it includes roughly 150 non-agricultural occupations, whereas the 1950 occupational classification in our linked sample has around 220 occupations.

29

The final panels in these figures instead assign individuals the most widely-used income score

in the literature: the IPUMS variable occscore (which is solely based on an individual’s

occupation). Relative to the other measures, this approach has the advantage that the income

assignment is symmetric for farmers and non-farmers. However, it is a very coarse measure as it

does not allow for an immigrant penalty in the first generation (which would tend to dampen

measured mobility). Yet, we still see that sons of immigrants can expect to reach a higher rank in

the earnings distribution than US-born individuals growing up with comparable income scores,

particularly in the 1880–1910 cohort.52

Overall, although the exact size of the intergenerational gaps varies as we consider different

proxies for individual-level income, in all cases we find higher levels of upward mobility among

children of immigrants.

C. Sensitivity of Results to Sample Construction

As noted above, we define a son’s immigrant status using his father’s birthplace. In Section 5

of the Online Appendix, we replicate all the main figures and tables using mother’s birthplace

instead and show that results remain unchanged.53 Among individuals who had at least one

foreign-born parent, two thirds had both parents born abroad. Our main finding is unchanged in

Appendix Figure A27, which defines a son’s immigrant status using both his mother’s and his

father’s birthplace (i.e., classifying sons as children of immigrants if both the mother and father

are born in one of the 17 countries of interest). Appendix Figure A28 separates the sample further

into sons with immigrant mothers only, immigrant fathers only, and two immigrant parents.

Although sample sizes get smaller, we find that the social mobility of all three groups of children

are above their counterparts with US-born parents.

In the past, immigrants were overwhelmingly white, and so we compare immigrants to the

white US-born population in the historical data. Today, in contrast, immigrants are more ethnically

diverse. Thus, our main specification for the modern period compares immigrants to the entire US-

born population. We note, however, that results are similar when comparing white immigrants to

the white population in the modern period, as we do in the historical data (panel (a) of Appendix

52 Appendix Figures A23 and A24 show that the cross-sectional results presented in Section 3 are robust to using the alternative income scores discussed in this section. 53 We show the main rank-rank correlations as well as the average ranks for children born to the 25th and 75th percentile of the income distribution by father’s country of origin in Appendix Figures A25 and A26, respectively.

30

Figure A.29), or all immigrants to the white population in the modern period (panel (b) of

Appendix Figure A.29).54

Finally, in Section 6 of the Online Appendix we explore how our past-present comparisons

differ when we construct the historical samples to match the characteristics of the Opportunity

Insights sample. More precisely, we restrict the analysis to sons ages 31–37, we loosen the

restriction on father’s age, and we use a measure of family income (as opposed to father’s) that

assigns predicted income to working mothers. Section 6 in the Online Appendix replicates the

main figures and tables using this alternative sample. The main findings of this exercise continue

to suggest that second-generation immigrants growing up at the 25th percentile of the income rank

end up 4–7 percentiles higher in the income rank than comparable children of US-born individuals.

7. Conclusions

We use newly-constructed father-son linked datasets to study whether the children of

immigrants achieve earnings parity with the children of the US-born, and how the intergenerational

mobility of immigrants has changed over the last two centuries. We find that, both in the past and

today, children of immigrants had greater chances of moving up in the income distribution relative

to the children of US-born parents with comparable family income or occupation score. Second

generation immigrants growing up in the 25th percentile end up 5–8 percentiles higher in the

income rank than the children of the US born. The children of immigrants achieve similar rates of

upward mobility despite large changes in country of origin (immigrants hailed from Europe in the

past, and from Latin America and Asia today), initial income levels (immigrants today starting

from lower levels), and US immigration policy over the last century (the ending of open borders

for European immigrants and the imposition of additional regulations for immigrant entry).

Our finding that children of immigrants have on average higher upward mobility is not driven

by immigrants from any particular ethnic origin. Rather, when estimating rates of intergenerational

mobility separately for each country of origin, we find that children of immigrants from nearly

every sending country have higher rates of upward mobility than the children of the US-born.

Using the historical data to dig deeper into why children of immigrants were more upwardly

mobile, we find that immigrant families were more likely than the US born to move to areas that

offered better prospects for their children. Furthermore, immigrant parents were “under-placed” in

54 As discussed above, we note that in this case we define second-generation status based on the mother’s birthplace.

31

the income distribution, thus allowing their children who were native English speakers and

educated in the US more scope for upward mobility.

The success of second-generation immigrants gives a more optimistic view of immigrant

assimilation than previous studies that have focused only on the first generation. Indeed, we find

that second generation immigrants overtake, rather than just catch up with the children of the US-

born with comparable family incomes. Although some politicians have a short-term perspective

on immigrant assimilation, our findings suggest that this view might underestimate the long-run

success of immigrants. Our findings are more consistent with the idea of the “American Dream,”

by which even immigrants who come to the United States with few resources and little skills have

a real chance at improving their children’s prospects.

32

References

Abramitzky, Ran, and Leah Boustan. 2017. “Immigration in American Economic History.” Journal of Economic Literature 55(4): 1311–1345. Abramitzky, Ran, Leah Boustan, Katherine Eriksson, James Feigenbaum, and Santiago Pérez. 2019. “Automated Linking of Historical Data.” Working Paper. Abramitzky, Ran, Leah Platt Boustan, and Katherine Eriksson. 2012. “Europe’s Tired, Poor, Huddled Masses: Self-Selection and Economic Outcomes in the Age of Mass Migration.” American Economic Review 102(5): 1832–1856. Abramitzky, Ran, Leah Platt Boustan, and Katherine Eriksson. 2014. “A Nation of Immigrants: Assimilation and Economic Outcomes in the Age of Mass Migration.” Journal of Political Economy 122(3): 467–506. Aydemir, Abdurrahman, Wen-Hao Chen, and Miles Corak. 2009. “Intergenerational Earnings Mobility Among the Children of Canadian Immigrants.” The Review of Economics and Statistics 91(2): 377–397. Bhattacharya, Debopam and Bhashkar Mazumder. 2011. “A Nonparametric Analysis of Black-White Differences in Intergenerational Income Mobility in the United States” Quantitative Economics 2: 335–379. Bleakley, Hoyt and Aimee Chin. 2010. “Age at Arrival, English Proficiency, and Social Assimilation among US Immigrants.” American Economic Journal: Applied Economics 2(1): 165–92. Bolotnyy, Valentin and Cristina Bratu. 2018. “The Intergenerational Mobility of Immigrants and the Native-Born: Evidence from Sweden.” Working Paper. Borjas, George. 1992. “Ethnic Capital and Intergenerational Mobility.” The Quarterly Journal of Economics 107(1): 123–150. Borjas, George. 1993. “The Intergenerational Mobility of Immigrants.” Journal of Labor Economics 11(1): 113–135. Cadena, Brian C. and Brian K. Kovak. “Immigrants Equilibrate Local Labor Markets: Evidence from the Great Recession.” American Economic Journal: Applied Economics 8(1): 257–290. Card, David, John DiNardo, and Eugenia Estes. 2000. “The More Things Change: Immigrants and the Children of Immigrants in the 1940s, the 1970s, and the 1990s.” in George J. Borjas (Ed.), Issues in the Economics of Immigration. Chicago: University of Chicago Press Card, David, Ciprian Domnisoru, and Lowell Taylor. 2018. “The Intergenerational Transmission of Human Capital: Evidence from the Golden Age of Upward Mobility.” Working Paper 25000, National Bureau of Economic Research.

33

Chetty, Raj, David Grusky, Nathaniel Hendren, Maximilian Hell, Robert Manduca, and Jimmy Narang. 2017. “The Fading American Dream: Trends in Absolute Income Mobility Since 1940.” Science. Chetty, Raj, John N. Friedman, Nathaniel Hendren, Maggie R. Jones, and Sonya R. Porter. 2018a. “Race and Economic Opportunity in the United States: An Intergenerational Perspective.” Working Paper 24441, National Bureau of Economic Research. Chetty, Raj, John N. Friedman, Nathaniel Hendren, Maggie R. Jones, and Sonya R. Porter. 2018b. “The Opportunity Atlas: Mapping the Childhood Roots of Social Mobility.” Working Paper 25147, National Bureau of Economic Research. Chetty, Raj, Nathaniel Hendren, Patrick Kline, and Emmanuel Saez. 2014. “Where is the land of Opportunity? The Geography of Intergenerational Mobility in the United States”. Quarterly Journal of Economics, 129(4): 1553–1623. Collins, William J. and Marianne H. Wanamaker. 2017. “Up from Slavery? African American Intergenerational Economic Mobility Since 1880.” Working Paper 23395, National Bureau of Economic Research. Collins, William J. and Ariell Zimran. 2019. “The Economic Assimilation of Irish Famine Migrants to the United States.” Working Paper 25287, National Bureau of Economic Research. Connor, Dylan Shane. 2019. “Class Background, Reception Context, and Intergenerational Mobility: A Record Linkage and Surname Analysis of the Children of Irish Immigrants.” International Migration Review. Derenoncourt, Ellora. 2019. “Can you move to opportunity? Evidence from the Great Migration.” Working Paper. Davis, Jonathan and Bhash Mazumder. 2019. “The Decline in Intergenerational Mobility After 1980.” Federal Reserve Bank of Chicago Working Paper No. 2017-05. Duncan, Brian, Jeffrey Grogger, Ana Sofia Leon and Stephen J. Trejo. 2017. “New Evidence of Generational Progress for Mexican Americans.” Working Paper 24067, National Bureau of Economic Research. Feigenbaum, James J. 2018. “Multiple Measures of Historical Intergenerational Mobility: Iowa 1915 to 1940.” Economic Journal 128(612): F446–F481. Feigenbaum, James J. 2015. “Intergenerational Mobility during the Great Depression.” Working Paper. Ferrie, Joseph P. 1997. “The Entry Into The US Labor Market Of Antebellum European Immigrants, 1840–1860.” Explorations in Economic History 31(3): 295–330.

34

Ferrie, Joseph, Catherine Massey, and Jonathan Rothbaum. 2016. “Do Grandparents and Great-Grandparents Matter? Multigenerational Mobility in the US, 1910–2013.” Working Paper 22635, National Bureau of Economic Research. Ferrie, Joseph and Jason Long. 2013. “Intergenerational Occupational Mobility in Great Britain and the United States since 1850.” American Economic Review 103(4): 1109–1137. Flood, Sarah, Miriam King, Renae Rodgers, Steven Ruggles, and J. Robert Warren. Integrated Public Use Microdata Series, Current Population Survey: Version 6.0 [dataset]. Minneapolis, MN: IPUMS, 2018. https://doi.org/10.18128/D030.V6.0. Gielen, Anne and Dinand Webbink. 2019. “Migration Selection and Assimilation: Evidence from the Great Migration from Surinam to the Netherlands” Working Paper. Goldin, Claudia. 1998. “America's Graduation from High School: The Evolution and Spread of Secondary Schooling in the Twentieth Century.” Journal of Economic History 58(2): 345–374. Goldenweiser, E. A. 1916. “The Farmer's Income." American Economic Review 6(1): 42–48. Hatton, Timothy. 1997. “The Immigrant Assimilation Puzzle in Late Nineteenth Century America.” Journal of Economic History 57(1): 34–62. Hilger, Nathaniel. 2016. “The Great Escape: Intergenerational Mobility in the United States Since 1940.” Working Paper 21217, National Bureau of Economic Research. Hilger, Nathaniel. 2017. “Upward Mobility and Discrimination: The Case of Asian Americans.” Working Paper 22748, National Bureau of Economic Research. Kasinitz, Philip, John H. Mollenkopf, Mary C. Waters and Jennifer Holdaway. 2009. Inheriting the City: The Children of Immigrants Come of Age. New York: Russell Sage Foundation. Kosack, Edward and Zachary Ward. 2018. “El sueño americano? The generational progress of Mexican Americans prior to World War II.” Working Paper. Lee, Chul-In and Gary Solon. 2009. “Trends in Intergenerational Income Mobility.” Review of Economics and Statistics 91(4): 766–772. Long, Jason and Joseph Ferrie. 2013. “Intergenerational Occupational Mobility in Great Britain and the United States since 1850.” American Economic Review 103(4): 1109–1137. Lubotsky, Darren. 2007. “Chutes or Ladders? A Longitudinal Analysis of Immigrant Earnings.” Journal of Political Economy 115(5): 820–867. Manson, Steven, Jonathan Schroeder, David Van Riper, and Steven Ruggles. 2019. IPUMS National Historical Geographic Information System: Version 14.0 [Database]. Minneapolis, MN: IPUMS. http://doi.org/10.18128/D050.V14.0

35

Mazumder, Bhashkar. 2005. “Fortunate Sons: New Estimates of Intergenerational Mobility in the United States Using Social Security Earnings Data.” The Review of Economics and Statistics 87(2): 235–255. Mazumder, Bhashkar. 2015. “Estimating the Intergenerational Elasticity and Rank Association in the US: Overcoming the Current Limitations of Tax Data.” Federal Reserve Bank of Chicago Working Paper No. 2015-04. Minns, Chris. 2000. “Income, Cohort Effects and Occupational Mobility: A New Look at Immigration to the United States at the Turn of the 20th Century.” Explorations Economic History 37(4): 326–50. Mitchell, Wesley, Willford I. King, Frederick R. Macaulay, and Oswald W. Knauth. 1922. “Income in the United States: Its Amount and Distribution, 1909–1919.” New York: National Bureau of Economic Research. Mazumder, Bhashkar. 2005. “Fortunate Sons: New Estimates of Intergenerational Mobility in the United States Using Social Security Earnings Data.” The Review of Economics and Statistics 87(2): 235–255. Olivetti, Claudia and M. Daniele Paserman. 2015. “In the Name of the Son (and the Daughter): Intergenerational Mobility in the United States, 1850-1940.” American Economic Review 105(8): 2695–2724. Olivetti, Claudia, M. Daniele Paserman, and Laura Salisbury. 2018. “Three-generation mobility in the United States, 1850–1940: The role of maternal and paternal grandparents.” Explorations in Economic History Vol. 70: 73–90. Perlmann, Joel. 2005. Italians Then, Mexicans Now: Immigrant Origins and Second-Generation Progress, 1890–2000. Russell Sage Foundation. Perlmann, Joel and Roger Waldinger. 1997. “Second generation decline? Children of immigrants, past and present--a reconsideration.” The International Migration Review 31(4): 893–922. Pérez, Santiago. 2019. “Intergenerational Occupational Mobility Across Three Continents.” The Journal of Economic History 79(2) 383–416. Preston, Samuel H., and Michael R. Haines. 1991. Fatal Years: Child Mortality in Late Nineteenth Century America. Princeton, NJ: Princeton University Press Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas, and Matthew Sobek. IPUMS USA: Version 8.0 [dataset]. Minneapolis, MN: IPUMS, 2018. https://doi.org/10.18128/D010.V8.0 Saavedra, Martin and Tate Twinam. 2018. “A Machine Learning Approach to Improving Occupational Income Scores.” Working Paper.

36

Tan, Hui Ren. 2018. “A Different Land of Opportunity: The Geography of Intergenerational Mobility in the Early 20th-Century US.” Working Paper. Ward, Zachary. Forthcoming. “The Not-So-Hot Melting Pot: The Persistence of Outcomes for Descendants of the Age of Mass Migration.” American Economic Journal: Applied Economics. . Ward, Zachary and Rohan Alexander. 2018. “Age at Arrival and Assimilation during the Age of Mass Migration.” The Journal of Economic History 78(3) 904–937. Ward, Zachary. 2019. “The low return to English fluency during the Age of Mass Migration.” Working Paper.

Figures

Figure 1: Cross-Sectional Earning Gaps for First- and Second-Generation Immigrants relative toUS-Born, by Country of Origin

(a) 1880–1910 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

d

Sweden

Russia

Italy

Denmark

Portug

al

Belgium

Wale

s

Canad

a

Austria

Switzerl

andIre

land

German

yFran

ce

Englan

d

Scotla

nd

First-generation Second-generation

(b) 1910–1940 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Norway

Portug

al

Austria

Denmark

Belgium

Sweden

German

yW

alesRuss

ia

Switzerl

and

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

First-generation Second-generation

(c) 1980–2010 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Vietna

mD.R.

MexicoHait

i

El Salv

ador

Jamaic

a

Ecuad

or

Guatem

ala

Colombia

South

KoreaCub

aChin

a

Greece

Italy

Philipp

inesIsr

ael

Canad

a

German

yInd

iaJap

anU.K

.

First-generation Second-generation

Note: The historical samples use the 1880, 1910, and 1940 Censuses. The historical samples of first-generationimmigrants and the US born are restricted to white men in the earlier Census who are aged 30–50, who have a childin the household, and who were either born in the US or in one of the 17 source countries. The second-generationsamples are restricted to white men in the later Census who are aged 30–50 and whose fathers were either born inthe US or in one of the 17 source countries. For the most recent cohort, we use the 1980 Census and the 2006–2015Current Population Survey (CPS)’s Annual Social and Economic Supplement. We restrict the CPS sample ofsecond-generation immigrants and their counterparts to men aged 30–50 whose fathers were either born in the USor in one of the top sending countries. We restrict the 1980 sample of first-generation immigrants and theircounterparts to men aged 30–50, who have a child present in the household, and who were born in the US or in oneof the top sending countries identified in the Opportunity Insights sample. For all of these cohorts, we use predictedincome as the measure of income.

1

Figure 2: Intergenerational Mobility of Immigrants and the US born, Rank-Rank Correlations

(a) 1880–1910 Cohort30

4050

6070

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 1910–to 1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) GSS: 1984–2006 Cohort

3040

5060

70So

ns ra

nk

0 20 40 60 80 100Father rank

US-born father Immigrant father

(d) Opportunity Insights: 1997–2015 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born father Immigrant father

Note: Sons are assigned percentile ranks relative to all other respondents born in the same birth year. Fathers areranked relative to all fathers with children in the same birth year. The figures plot the mean income rank ofchildren by father’s income rank, for sons with and without foreign-born fathers, as well as the correspondingregression lines using equation (2). In the GSS sample, the outcomes for sons are measured between 2000 and 2018(with the median year being 2006). The outcomes for parents are measured via a retrospective question aboutparents’ occupations when the respondent was 16 years old; the data therefore correspond to parental outcomesbetween 1966 and 2004 (with the median year being 1984). The Opportunity Insights Data comes from Chetty etal. (2018a); son’s income is measured in 2014–2015 and parental income is measured in 1994–2000.

2

Figure 3: Average Income Rank for Children Born to 25th Percentile, by Father’s Birthplace

(a) 1880–1910 Cohort25

4565

Ave

rage

inco

me

rank

Norway

Belgium U.S.

German

y

Switzerl

and

SwedenFran

ce

Austria

Denmark

Englan

d

Scotla

ndW

ales

Canad

aRuss

ia

Irelan

d

Finlan

dIta

ly

Portug

al

(b) 1910–1940 Cohort

2545

65

Ave

rage

inco

me

rank

Norway

German

yU.S.

Switzerl

and

Denmark

Sweden

Finlan

d

Belgium

Austria

Canad

aFran

ceW

ales

Englan

d

Scotla

ndRuss

ia

Portug

al

Irelan

dIta

ly

(c) Opportunity Insights: 1997–2015 Cohort

2545

65

Ave

rage

inco

me

rank

Jamaic

aHait

iU.S.

U.K.

German

y

MexicoD.R.

Japan

Cuba

Canad

aIsr

ael

El Salv

ador

Philipp

ines

Guatem

ala

Ecuad

or

Colombia

South

Korea

Greece

Italy

Vietna

mInd

iaChin

a

Note: This figure plots the average income rank for children born to the 25th percentile of the parent incomedistribution. Data for the Opportunity Insights cohort come from Chetty et al. (2018a); son’s income is measuredin 2014–2015 and parental income is measured in 1994–2000.

3

Figure 4: Average Income Rank for Children Born to 75th Percentile, by Father’s Birthplace

(a) 1880–1910 Cohort35

5575

Ave

rage

inco

me

rank

Belgium

Norway

Switzerl

and

Portug

al

Sweden

Wale

sU.S.

Denmark

France

Canad

a

German

y

Scotla

nd

Englan

d

Finlan

dIre

land

Austria Ita

lyRuss

ia

(b) 1910–1940 Cohort

3555

75

Ave

rage

inco

me

rank

Finlan

d

Portug

al

Belgium U.S.

Norway

Canad

a

Switzerl

and

Denmark

German

yW

ales

Sweden Ita

ly

Englan

dFran

ce

Scotla

ndIre

land

Austria

Russia

(c) Opportunity Insights: 1997–2015 Cohort

3555

75

Ave

rage

inco

me

rank

Haiti

Jamaic

aD.R.

Guatem

ala

El Salv

ador

Mexico

Japan

U.K.

German

yU.S.

Ecuad

or

Philipp

ines

Canad

a

ColombiaCub

aIsr

ael

South

Korea

Vietna

mGree

ceIta

lyChin

aInd

ia

Note: This figure plots the average income rank for children born to the 75th percentile of the parent incomedistribution. Data for the Opportunity Insights cohort come from Chetty et al. (2018a); son’s income is measuredin 2014–2015 and parental income is measured in 1994–2000.

4

Figure 5: Intergenerational Mobility of Immigrants and the US born, Rank-Rank Correlations byImmigrant Group (Modern Cohort)

(a) White

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born mother Immigrant mother

(b) Hispanic

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born mother Immigrant mother

(c) Asian

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born mother Immigrant mother

Note: Data come from Chetty et al. (2018b). Sons are assigned percentile ranks relative to all other respondentsborn in the same birth year. Parents are also ranked relative to all fathers with children in the same birth year. Thefigures plot the mean income rank of children by their household’s income rank, for sons with and withoutforeign-born mothers, as well as the corresponding regression lines. The figures include individuals of all racesamong those with a US-born mother, and individuals of the specified race among those with a non-US-born mother.

5

Figure 6: Intergenerational Gap at the 25th percentile for 1910–1940 Cohort, by Father’s Age ofArrival

0.51

2.17

5.28

02

46

810

Gap

at 2

5th

perc

entil

e

7 or younger Age 8-16 17 or older

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 25th percentile of the income distribution. Sons of immigrants are divided into groups based ontheir father’s age of arrival to the United States.

Figure 7: Intergenerational Gap at the 25th percentile for 1910–1940 Cohort, by Father’s Age ofArrival and Language in Sending Country

(a) Non-English-Speaking Countries

1.12

4.68

-0.97

-20

24

68

10

Gap

at 2

5th

perc

entil

e

7 or younger Age 8-16 17 or older

(b) English-Speaking Countries

5.91

6.87

8.67

-20

24

68

10

Gap

at 2

5th

perc

entil

e

7 or younger Age 8-16 17 or older

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 25th percentile of the income distribution. The figures first separate sons based on thelanguage spoken in the father’s sending country (where English-speaking countries are Canada, England, Ireland,Scotland, and Wales). Then, sons of immigrants are divided into groups based on their father’s age of arrival to theUnited States.

6

Figure 8: Educational Attainment, 1910–1940 cohort

(a) Son graduated high school.1

.2.3

.4.5

.6

Gra

duat

ed h

igh

scho

ol

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) Son’s highest grade in school

89

1011

12

Hig

hest

gra

de a

ttain

ed

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) Son’s school attendance in 1910

.7.8

.91

Atte

ndin

g sc

hool

in c

hild

hood

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(d) Son’s income rank and educational attainment

020

4060

80

Son

rank

0 3 6 9 12 15 18

Highest grade attained

US-born father Immigrant father

(e) Son’s and father’s educational attainment

68

1012

14

Son

high

est g

rade

atta

ined

0 3 6 9 12 15 18

Father highest grade attained

US-born father Immigrant father

Note: Fathers are ranked relative to all fathers with children in the same birth year. Panels (a) and (b) plot themean educational attainment of children by their father’s income rank, for sons with and without foreign-bornfathers, as well as the corresponding regression lines. Panel (c) considers linked sons aged 12–16 in the earlierCensus and plots school attendance as a function of his father’s income rank (after controlling for age fixed effects).Panel (d) plots a son’s mean income rank as a function of completed years of schooling. Panel (e) plots a son’smean years of completed schooling as a function of father’s years of schooling (for fathers who can be linked to the1940 Census).

7

Figure 9: Share of Immigrants and Upward Mobility, by State Economic Area or Commuting Zone

(a) Share of sons that are second-generation(1880–1910)

(b) Upward mobility of sons of US-born fathers(1880–1910)

(c) Share of sons that are second-generation(1910–1940)

(d) Upward mobility of sons of US-born fathers(1910–1940)

(e) Share of population that is foreign-born, 1980 (f) Upward mobility of Opp. Insights cohort

Note: Panels (a) and (c) show the share of the individuals in our linked sample that are second-generation immigrants in each1880 or 1910 state economic area, respectively. In panels (b) and (d), upward mobility is measured as the share of sons in eachstate economic area that reached the top third of the national income distribution, conditional on having had a father in thebottom third of the national income distribution. Panel (e) shows the share each of each 1980 county’s population that isforeign-born. Panel (f) shows the share of white sons who are upwardly mobile using data from Chetty et al. (2018b); upwardmobility is measured as the share of sons in the top 20 percent of the income distribution, conditional on having hadhousehold income in the 25th percentile of the income distribution.

8

Figure 10: Share of Immigrants and Upward Mobility, by State Economic Area or County

(a) 1880–1910 Cohort0

.2.4

.6.8

1

Shar

e up

war

dly

mob

ile, U

S-bo

rn fa

ther

s

0 .2 .4 .6 .8 1

Share second-generation immigrants

Northeast Midwest West South

(b) 1910–1940 Cohort

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile, U

S-bo

rn fa

ther

s

0 .2 .4 .6 .8

Share second-generation immigrants

Northeast Midwest West South

(c) 1980 Census and Opp. Insights Cohort

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile

0 .1 .2 .3 .4

Share foreign-born

Northeast Midwest West South

Note: The first two panels use the linked samples to plot the share of a childhood SEA population that is a second-generation immigrant against thecorresponding share of children with US-born fathers who were upwardly mobile. Upward mobility is measured as the share of sons in each SEA that reachedthe top third of the national income distribution, conditional on having had a father in the bottom third of the corresponding national income distribution. Thethird panel plots the share of foreign-born individuals in each 1980 county against the share of sons that were upwardly mobile in that county. Data on upwardmobility come from Chetty et al. (2018b); mobility is measured as the share of white sons in the top 20 percent of the income distribution, conditional on havinghad household income in the 25th percentile of the income distribution. Data on county-level foreign-born shares from the 1980 Census come from Manson et al.(2019). Each graph includes a line of best fit.

9

Figure 11: Intergenerational Gap at the 25th percentile, Comparing Children in Similar ChildhoodLocations

(a) 1880–1910 Cohort

3.41

0.96

0.09

-1.36 -1.57

-4-2

02

46

8

Gap

at t

he 2

5th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

(b) 1910–1940 Cohort

3.99

1.070.58

-1.32 -1.27

-4-2

02

46

8

Gap

at t

he 2

5th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

(c) GSS: 1984–2006 Cohort

4.63 4.63

02

46

810

Gap

at 2

5th

perc

entil

e

Father

Incom

e

+ Childh

ood R

egion

FE

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 25th percentile of the income distribution. The first bar for each cohort is estimated byregressing son’s income ranks on their fathers’ ranks, an indicator for having a foreign-born father, and theinteraction of these two variables. The second bar for each cohort plots the gap after including childhood regionfixed effects. In the historical cohorts, the third bar plots the gap after including childhood state fixed effects andthe fourth bar plots the gap after including state × urban fixed effects. Finally, the fifth bar plots the gap afterincluding childhood county fixed effects.

10

Tables

Table 1: Intergenerational Mobility Estimates, by Cohort

Intercept α Slope β

US-bornfather

Immigrantfather T-statistic

US-bornfather

Immigrantfather T-statistic

1880–1910 Cohort 31.34 35.68 36.49 0.36 0.32 19.35

1910–1940 Cohort 31.00 34.67 43.57 0.36 0.37 9.16

GSS Cohort 35.42 42.18 2.52 0.29 0.21 1.58

Opp. Insights Cohort 36.64 44.05 12.46 0.33 0.25 8.13

Note: This table reports the slope and intercept from regressions of son’s rank on father’s rank, by cohort. The third and sixth columns correspond to the testof the hypothesis that both estimates are equal to each other. Data for the Opportunity Insights cohort come from Chetty et al. (2018a).

11

Appendix Figures and Tables

Figure A1: Average Earnings Gaps Using Actual Income, by Sending Country and Generation

(a) 1910–1940 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Norway

Portug

al

Austria

Denmark

Belgium

Sweden

German

yW

alesRuss

ia

Switzerl

and

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

First-generation Second-generation

(b) 1980–2010 Cohort

-.8-.6

-.4-.2

0.2

.4.6

.81

Logg

ed in

com

e sc

ore

Vietna

mD.R.

MexicoHait

i

El Salv

ador

Jamaic

a

Ecuad

or

Guatem

ala

Colombia

South

KoreaCub

aChin

a

Greece

Italy

Philipp

inesIsr

ael

Canad

a

German

yInd

iaJap

anU.K

.

First-generation Second-generation

Note: Similar to figure 1, the historical sample uses the 1910 and 1940 Censuses and the modern cohort uses the1980 Census and the 2006–2015 Current Population Survey (CPS)’s Annual Social and Economic Supplement.However, instead of using predicted income, we use income from wages (adjusted for self-employed and farmerincome) for the 1940 sons and total family income for the fathers and sons in the modern cohort.

12

Figure A2: Cross-Sectional Earnings Gaps for First-Generation Fathers and Second-GenerationDaughters relative to US-Born, by Country of Origin

(a) 1880–1910 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

d

Sweden

Russia

Italy

Denmark

Portug

al

Belgium

Wale

s

Canad

a

Austria

Switzerl

andIre

land

German

yFran

ce

Englan

d

Scotla

nd

First-generation Second-generation

(b) 1910–1940 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Norway

Portug

al

Austria

Denmark

Belgium

Sweden

German

yW

alesRuss

ia

Switzerl

and

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

First-generation Second-generation

(c) 1980–2010 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Vietna

mD.R.

MexicoHait

i

El Salv

ador

Jamaic

a

Ecuad

or

Guatem

ala

Colombia

South

KoreaCub

aChin

a

Greece

Italy

Philipp

inesIsr

ael

Canad

a

German

yInd

iaJap

anU.K

.

First-generation Second-generation

Note: The historical samples use the 1880, 1910, and 1940 Censuses. The historical samples of first-generationimmigrants and their US-born counterparts are white men in the earlier Census who are aged 30–50, who have achild in the household, and who were either born in the US or in one of the 17 source countries. The historicalsamples of second-generation immigrants and their counterparts are US-born white women in the later Census whoare aged 30–50 and whose fathers were either born in the US or in one of the 17 source countries. For the mostrecent cohort, we use the 1980 Census and the 2006–2015 Current Population Survey (CPS)’s Annual Social andEconomic Supplement. We restrict the CPS sample of second-generation immigrants and their counterparts toUS-born women who are aged 30–50 and whose fathers were either born in the US or in one of the top sendingcountries. We restrict the 1980 sample of first-generation immigrants and their counterparts to men who are aged30–50, who have a child present in the household, and who were born in the US or in one of the top sendingcountries identified in the CPS sample. For all of these cohorts, we use predicted income as the measure of income;for the second generation, we calculate predicted income for a woman as the sum of her income (if she is employed)and her spouse’s income (if she is married).

13

Figure A3: Average Earnings Gaps in Linked Sample, by Sending Country and Generation

(a) 1880–1910 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Sweden

Finlan

d

Belgium

Denmark Ita

lyW

alesRuss

ia

Canad

a

Portug

al

Switzerl

and

Austria

Irelan

dFran

ce

Englan

d

German

y

Scotla

nd

Fathers Sons

(b) 1910–1940 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

dIta

ly

Denmark

Portug

al

Belgium

Austria

Sweden

German

y

Switzerl

andW

alesRuss

ia

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

Fathers Sons

Note: Both of these figures use our baseline linked samples, classifying immigrant sons using the father’s birthplace.For all individuals, we use our measure of predicted income in conjunction with CW adjustments for farmer income.

Figure A4: Average Earnings Gaps Excluding US South in Linked Sample, by Sending Countryand Generation

(a) 1880–1910 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Sweden

Finlan

d

Belgium

Denmark

Russia

Wale

sIta

ly

Canad

a

Portug

al

Switzerl

and

Austria

France

Irelan

d

Englan

d

German

y

Scotla

nd

Fathers Sons

(b) 1910–1940 Cohort

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

dIta

ly

Denmark

Portug

al

Belgium

Austria

Sweden

German

y

Switzerl

andW

alesRuss

ia

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

Fathers Sons

Note: Both of these figures use our baseline linked samples, classifying immigrant sons using the father’sbirthplace. We exclude father-son pairs who lived in the South in either the earlier or later Census. For allindividuals, we use our measure of predicted income in conjunction with CW adjustments for farmer income.

14

Figure A5: Rank-rank Correlations, by Grandfather’s Birthplace

(a) 1880–1910 Cohort: Rank-rank

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father, US-born grandfather US-born father, imm. grandfatherImmigrant father & grandfather

(b) 1910–1940 Cohort: Rank-rank

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father, US-born grandfather US-born father, imm. grandfatherImmigrant father & grandfather

Note: This figure divides sons with US-born fathers into two groups based on their grandfather’s place of birth:those whose father’s father was born in the US and those whose father’s father was born abroad. Sons are assignedpercentile ranks relative to all other respondents born in the same birth year. Fathers are also ranked relative to allfathers with children in the same birth year. The figures plot the mean income rank of children by father’s incomerank for each group as well as the corresponding regression lines.

Figure A6: Rank-rank Correlations Using Actual Income

(a) 1910–1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) GSS Cohort

3040

5060

7080

90So

ns ra

nk

0 20 40 60 80 100Father rank

US-born father Immigrant father

Note: This figure uses actual income for the 1940 sons (adjusted for self-employed and farm income) and for sonsin the GSS cohort. Sons are assigned percentile ranks relative to all other respondents born in the same birth year.Fathers are also ranked relative to all fathers with children in the same birth year. The figures plot the meanincome rank of children by father’s income rank, for sons with and without foreign-born fathers, as well as thecorresponding regression lines.

15

Figure A7: Weighted Average Income Rank for Children Born to 25th Percentile, by Cohort andFather’s Birthplace

(a) 1880–1910 Cohort

2545

65

Ave

rage

inco

me

rank

Norway

Belgium U.S.

German

y

Switzerl

and

SwedenFran

ce

Denmark

Austria

Englan

d

Scotla

ndW

ales

Canad

a

Finlan

dIre

landRuss

iaIta

ly

Portug

al

(b) 1910–1940 Cohort

2545

65

Ave

rage

inco

me

rank

Norway

German

y

Finlan

dU.S.

Switzerl

and

Belgium

Austria

Sweden

Denmark

France

Canad

aW

ales

Englan

dRuss

ia

Scotla

nd

Portug

al

Irelan

dIta

ly

Note: This figure plots the average income rank for children born to the 25th percentile of the parent incomedistribution. Each observation is weighted by the inverse probability of being a linked individual using theindividual’s childhood characteristics.

16

Mechanisms: Additional Results

Figure A8: Rank-rank Correlations for 1910–1940 cohort, by Father’s Age at Arrival

(a) 7 or younger

3040

5060

70

Son

inco

me

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) Ages 8–16

3040

5060

70

Son

inco

me

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) 17 or older

3040

5060

70

Son

inco

me

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: Sons are assigned percentile ranks relative to all other respondents born in the same birth year. Fathers areranked relative to all fathers with children in the same birth year. The figures plot the mean income rank of childrenby father’s income rank, for sons with and without foreign-born fathers, as well as the corresponding regressionlines. Sons of immigrants are divided into groups based on their father’s age of arrival to the United States.

17

Figure A9: Upward Mobility for Historical Cohorts, by State Economic Area and Father’s Birthplace

(a) Sons of US-born fathers (1880–1910) (b) Sons of immigrant fathers (1880–1910)

(c) Sons of US-born fathers (1910–1940) (d) Sons of immigrant fathers (1910–1940)

Note: Upward mobility is measured by the share of sons in each state that reached the top third of the national income distribution, conditional on having hada father in the bottom third of the corresponding national income distribution.

18

Figure A10: Share of Immigrants and Upward Mobility, by Locality

(a) 1880–1910 Cohort, State0

.2.4

.6.8

1

Shar

e up

war

dly

mob

ile, U

S-bo

rn fa

ther

s

0 .2 .4 .6 .8

Share second-generation immigrants

Northeast Midwest West South

(b) 1910–1940 Cohort, State

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile, U

S-bo

rn fa

ther

s

0 .2 .4 .6

Share second-generation immigrants

Northeast Midwest West South

(c) 1980 Census and Opp. Insights Cohort, CommutingZone

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile

0 10 20 30

Share first-generation immigrants

Northeast Midwest West South

Note: The first two panels of this graph use the linked samples to plot the share of a childhood state’s population that is a second-generation immigrantagainst the corresponding share of children with US-born fathers who are upwardly mobile. Upward mobility is measured by the share of sons in each state thatreached the top third of the national income distribution, conditional on having had a father in the bottom third of the national income distribution. Mobilitydata in the modern cohort come from Chetty et al. (2018b); upward mobility is measured as the share of white sons in the top 20 percent of the incomedistribution, conditional on having had household income in the 25th percentile of the income distribution. Data on immigrant shares comes from the 1980Census and is calculated as the share of men ages 30–50 who are foreign-born. Each graph includes a line of best fit.

19

Figure A11: Upward Mobility of Sons of Immigrant and US-born Men, by SEA or State

(a) 1880–1910 Cohort, SEA

Immmigrant Share of RegionNE = 33.44 MW = 33.32

S = 8.91 W = 40.56

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile, i

mm

igra

nt fa

ther

s

0 .2 .4 .6 .8 1

Share upwardly mobile, US-born fathers

Northeast Midwest West South

(b) 1880–1910 Cohort, State

Immmigrant Share of RegionNE = 33.44 MW = 33.32

S = 8.90 W = 40.56

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile, i

mm

igra

nt fa

ther

s

0 .2 .4 .6 .8 1

Share upwardly mobile, US-born fathers

Northeast Midwest West South

(c) 1910–1940 Cohort, SEA

Immmigrant Share of RegionNE = 39.97 MW = 25.64

S = 4.70 W = 28.73

0.2

.4.6

.81

Shar

e up

war

dly

mob

ile, i

mm

igra

nt fa

ther

s

0 .2 .4 .6 .8 1

Share upwardly mobile, US-born fathers

Northeast Midwest West South

(d) 1910–1940 Cohort, State

Immmigrant Share of RegionNE = 39.97 MW = 25.64

S = 4.70 W = 28.73 0

.2.4

.6.8

1

Shar

e up

war

dly

mob

ile, i

mm

igra

nt fa

ther

s

0 .2 .4 .6 .8 1

Share upwardly mobile, US-born fathers

Northeast Midwest West South

Note: Upward mobility is measured by the share of sons in each state that reached the top third of the nationalincome distribution, conditional on having had a father in the bottom third of the corresponding national incomedistribution. Each graph includes the 45 degree line.

20

Figure A12: Intergenerational Gap, with Location Fixed Effects

(a) 1880–1910: Gap at 50th percentile

2.47

1.02

0.14

-1.60 -1.64

-4-2

02

46

8

Gap

at t

he 5

0th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

(b) 1880–1910: Gap at 75th percentile

1.541.08

0.19

-1.85 -1.71

-4-2

02

46

8

Gap

at t

he 7

5th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

(c) 1910–1940: Gap at 50th percentile

4.31

1.80

0.88

-1.03 -1.13

-4-2

02

46

8

Gap

at t

he 5

0th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

(d) 1910–1940: Gap at 75th percentile

4.63

2.53

1.18

-0.74 -0.99

-4-2

02

46

8

Gap

at t

he 7

5th

perc

entil

e

Father

Incom

e

+ Child R

egion

FE

+ Child S

tate F

E

+Child S

tate-U

rban F

E

+Child C

ounty

FE

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 50th and 75th percentile of the income distribution. The first bar for each cohort is estimatedby regressing son’s income ranks on their fathers’ ranks, an indicator for having a foreign-born father, and theinteraction of these two variables. The second bar for each cohort plots the gap after including childhood regionfixed effects. The third bar plots the gap after including childhood state fixed effects and the fourth bar plots thegap after including state × urban fixed effects. Finally, the fifth bar plots the gap after including childhood countyfixed effects.

21

Figure A13: Intergenerational Gap Including County Fixed Effects, with Alternative IncomeMeasures for 1910–1940 Cohort

3.99

2.90

0.45

-1.27 -1.14-1.54

-4-2

02

46

8

Inte

rgen

erat

iona

l gap

Father

Incom

e

+Child C

ounty

FE

Father

Incom

e

+Child C

ounty

FE

Father

Incom

e

+Child C

ounty

FE

Predicted income Actual IncomeOccscore

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 25th percentile of the income distribution. For each income measure, the first bar in eachgroup shows the intergenerational gap (estimated by regressing sons’ income ranks on their fathers’ ranks, anindicator for having a foreign-born father, and the interaction of these two variables). The second bar plots the gapafter including childhood county fixed effects. “Predicted income” refers to our baseline measure of income for 1940sons. “Actual income” refers to using the 1940 variable for income from wages for the sons. “Occscore” refers tousing the IPUMS occupational score for the sons.

Figure A14: Intergenerational Gap, by Enclave Status

(a) 1880–1910 Cohort

4.05

1.17

2.87

-0.83

-20

24

68

10

Gap

at 2

5th

perc

entil

e

Not in

Enclav

e

+State F

E

Enclav

e

+State F

E

(b) 1910–1940 Cohort

5.12

2.07

3.14

-0.60

-20

24

68

10

Gap

at 2

5th

perc

entil

e

Not in

Enclav

e

+State F

E

Enclav

e

+State F

E

Note: This graph plots the intergenerational gap between children of immigrants and children of US-born men, forchildren born to the 25th percentile of the income distribution. Sons of immigrants are divided into two groupsbased on whether they grew up an immigrant enclave. An individual is classified as growing up in an enclave if atleast 10 percent of his childhood enumeration district’s adult population was from his father’s same country oforigin (calculated using the 1880 and 1910 cross-sectional Censuses, respectively). For each group, we show theintergenerational gap before and after controlling for childhood state FE.

22

Figure A15: Rank-rank Correlations, by US-born Father’s Migrant Status

(a) 1880–1910 Cohort: Rank-rank30

4050

6070

Son

rank

0 20 40 60 80 100

Father rank

US-born father, non-migrant US-born father, migrantImmigrant father

(b) 1910–1940 Cohort: Rank-rank

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father, non-migrant US-born father, migrantImmigrant father

Note: This figure divides sons with US-born men into two groups based on the father’s migrant status: thosewhose fathers internally migrated (i.e., their state of residence differs from their state of birth) and those who didnot internally migrate. Sons are assigned percentile ranks relative to all other respondents born in the same birthyear. Fathers are also ranked relative to all fathers with children in the same birth year. The figures plot the meanincome rank of children by father’s income rank for each group as well as the corresponding regression lines.

Figure A16: Likelihood of Moving from Childhood County

(a) 1880–1910 Cohort

.4.5

.6.7

.8

Mig

rate

d co

unty

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 1910–1940 Cohort

.4.5

.6.7

.8

Mig

rate

d co

unty

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: Sons are classified as having moved if the county that they reside in in the later Census is different from thecounty where they lived in the earlier Census. Fathers are ranked relative to all fathers with children in the samebirth year. The figures plot the likelihood of moving by father’s income rank, for children with and without aforeign-born father, as well as the corresponding regression lines.

23

Robustness to Linking Procedure

Figure A17: Rank-rank Correlations for 1880–1910 Cohort with Alternative Linking Strategies

(a) Exact names

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 5-year band

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) Exact names and 5-year band

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: Sons are assigned percentile ranks relative to all other respondents born in the same birth year. Fathers arealso ranked relative to all fathers with children in the same birth year. The figures plot the mean income rank ofchildren by father’s income rank, for sons with and without foreign-born fathers, as well as the correspondingregression lines. Each panel shows the results for the rank-based specification using alternative strategies for linkingfathers and sons across the 1880 and 1910 Censuses.

24

Figure A18: Rank-rank Correlations for 1910–1940 Cohort with Alternative Linking Strategies

(a) Exact names30

4050

6070

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 5-year band

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) Exact names and 5-year band

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: Sons are assigned percentile ranks relative to all other respondents born in the same birth year. Fathers arealso ranked relative to all fathers with children in the same birth year. The figures plot the mean income rank ofchildren by father’s income rank, for sons with and without foreign-born fathers, as well as the correspondingregression lines. Each panel shows the results for the rank-based specification using alternative strategies for linkingfathers and sons across the 1910 and 1940 Censuses.

25

Figure A19: Rank-rank Correlations, Weighted Data

(a) 1880–1910 Cohort, 1880 weight

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 1880–1910 Cohort, 1910 weight

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) 1910–1940 Cohort, 1910 weight

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(d) 1910–1940 Cohort, 1940 weight

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: This figure weights observations by the inverse probability of being a linked individual. Each cohort has twopossible weights: one that uses childhood characteristics and the earlier cross-sectional Census, and one that usesadult characteristics and the later cross-sectional Census. Sons are assigned percentile ranks relative to all otherrespondents born in the same birth year. Fathers are also ranked relative to all fathers with children in the samebirth year. The figures plot the mean income rank of children by father’s income rank, for sons with and withoutforeign-born fathers, as well as the corresponding regression lines. For more details on the construction of theweights, see Section 3 in the Online Appendix.

26

Robustness to Alternative Income Measures

Figure A20: Rank-rank Correlations for 1880–1910 Cohort Using Alternative Income Measures

(a) 1900 Census of Agriculture

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) No farmer fathers or sons

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) 1901 Cost of Living, 1900 Census of Ag.

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(d) Fathers, No Immigrant Penalty

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(e) Sons, Immigrant Penalty

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(f) IPUMS 1950 Occscore

3040

5060

70

Son

rank

20 40 60 80 100

Father rank

US-born father Immigrant father

Note: The figures plot the mean income rank of children by father’s income rank, for sons with and without foreign-born fathers, as well as the correspondingregression lines. Panel (a) keeps predicted income for non-farming occupations, but adjust farmers’ income using county-level measures derived from the 1900 Census ofAgriculture. Panel (b) drops fathers and sons who are farmers from the sample and re-runs the specification using ranks from the non-farming income distribution.Panel (c) keeps the farmer adjustments from panel (a), and also adjusts non-farmer income using the average earnings in a person’s occupation from the 1901 Cost ofLiving Survey. Panel (d) adjusts father’s predicted income to not incorporate their country of origin in the prediction. Panel (e) adjusts son’s predicted income toincorporate the father’s country of origin into the prediction (or father’s immigrant status for farmers). Panel (f) uses the 1950 IPUMS occscore measure.

Figure A21: Rank-rank Correlations for 1910–1940 Cohort Using Alternative Income Measures

(a) 1900 Census of Agriculture30

4050

6070

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) No farmer fathers or sons

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(c) 1901 Cost of Living, 1900 Census of Ag.

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(d) Fathers, No Immigrant Penalty

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(e) Sons, Immigrant Penalty

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(f) IPUMS 1950 Occscore

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: The figures plot the mean income rank of children by father’s income rank, for sons with and without foreign-born fathers, as well as the correspondingregression lines. Panel (a) keeps predicted income for non-farming occupations, but adjust farmers’ income using county-level measures derived from the 1900 Census ofAgriculture. Panel (b) drops fathers and sons who are farmers from the sample and re-runs the specification using ranks from the non-farming income distribution.Panel (c) keeps the farmer adjustments from panel (a), and also adjusts non-farmer income for fathers using the average earnings in a person’s occupation from the1901 Cost of Living Survey. Panel (d) adjusts father’s predicted income to not incorporate their country of origin in the prediction. Panel (e) adjusts son’s predictedincome to incorporate the father’s country of origin into the prediction (or father’s immigrant status for farmers). Panel (f) uses the 1950 IPUMS occscore measure.

Figure A22: Rank-rank Correlations Including Missing Income

(a) 1880–1910 Cohort30

4050

6070

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

(b) 1910–1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born father Immigrant father

Note: This figure assigns zero income to fathers and sons with missing occupation or income. Sons are assignedpercentile ranks relative to all other respondents born in the same birth year. Fathers are also ranked relative to allfathers with children in the same birth year. The figures plot the mean income rank of children by father’s incomerank, for sons with and without foreign-born fathers, as well as the corresponding regression lines.

Figure A23: Country-by-Country Estimates for 1880–1910 Cohort Using Alternative IncomeMeasures

(a) 1900 Census of Agriculture

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Belgium

Canad

a

Portug

al

Irelan

dW

ales

Austria

France

Denmark

German

y

Switzerl

and

Englan

d

Sweden

Scotla

nd

Norway

Russia

Fathers Sons

(b) No farmer fathers or sons

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

d

Portug

alIta

lyW

ales

Irelan

d

Canad

a

Norway

Belgium

Sweden

Denmark

Austria

France

German

y

Switzerl

and

Englan

dRuss

ia

Scotla

nd

Fathers Sons

(c) 1901 Cost of Living, 1900 Census of Ag.

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIre

land

Belgium

Canad

aW

ales

Portug

al

Englan

dIta

ly

German

y

Scotla

nd

Switzerl

and

Denmark

France

Sweden

Austria

Norway

Russia

Fathers Sons

(d) 1st generation, no Immigrant Penalty

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Sweden

Belgium

Finlan

dIre

landW

ales

Denmark

Switzerl

and

Canad

aFran

ce

Englan

d

German

yRuss

ia

Scotla

nd

Austria

Portug

alIta

ly

Fathers Sons

(e) 2nd generation, Immigrant Penalty

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Sweden

Finlan

d

Belgium

Denmark Ita

lyW

alesRuss

ia

Canad

a

Portug

al

Switzerl

and

Austria

Irelan

dFran

ce

Englan

d

German

y

Scotla

nd

Fathers Sons

(f) IPUMS 1950 Occscore

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Belgium

Sweden

Finlan

d

Denmark

Portug

al

Switzerl

and

Canad

aFran

ce

Englan

dIre

landW

alesRuss

ia

Scotla

nd

German

y

Austria Ita

ly

Fathers Sons

Note: All of these figures use our baseline linked samples, classifying immigrant families using the father’sbirthplace. Each panel plots the country-of-origin coefficients from equation (1), using alternative income measuresfor the fathers and sons. Panel (a) keeps predicted income for non-farming occupations, but adjust farmers’ incomeusing county-level measures derived from the 1900 Census of Agriculture. Panel (b) drops fathers and sons who arefarmers from the sample. Panel (c) keeps the farmer adjustments from panel (a), and also adjusts non-farmerincome using the average earnings in a person’s occupation from the 1901 Cost of Living Survey. Panel (d) adjustsson’s predicted income to incorporate their father’s country of origin into the prediction. Panel (e) adjusts father’spredicted income to not incorporate their country of origin in the prediction. Panel (f) uses the 1950 IPUMSoccscore measure.

Figure A24: Country-by-Country Estimates for 1910–1940 Cohort Using Alternative IncomeMeasures

(a) 1900 Census of Agriculture

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Portug

al

Austria

Belgium

Canad

aW

ales

Irelan

dFran

ceRuss

ia

Sweden

Norway

German

y

Switzerl

and

Englan

d

Scotla

nd

Denmark

Fathers Sons

(b) No farmer fathers or sons

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Portug

al

Finlan

dIta

ly

Austria

Belgium

Norway

Wale

s

Canad

aFran

ce

Irelan

dRuss

ia

Denmark

German

y

Sweden

Switzerl

and

Englan

d

Scotla

nd

Fathers Sons

(c) 1901 Cost of Living, 1900 Census of Ag.

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Finlan

dIta

ly

Belgium

Irelan

d

Portug

al

Austria

Wale

s

Canad

a

Englan

dFran

ce

Sweden

Scotla

nd

German

y

Switzerl

andRuss

ia

Norway

Denmark

Fathers Sons

(d) 1st generation, no Immigrant Penalty

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

d

Belgium

Denmark

Sweden

German

y

Austria

Wale

sIta

ly

Switzerl

and

Canad

a

Irelan

dFran

ce

Portug

al

Scotla

nd

Englan

dRuss

ia

Fathers Sons

(e) 2nd generation, Immigrant Penalty

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Norway

Finlan

dIta

ly

Denmark

Portug

al

Belgium

Austria

Sweden

German

y

Switzerl

andW

alesRuss

ia

Canad

aFran

ce

Irelan

d

Englan

d

Scotla

nd

Fathers Sons

(f) IPUMS 1950 Occscore

-.8-.6

-.4-.2

0.2

.4.6

Logg

ed in

com

e sc

ore

Portug

al

Norway

Denmark

Switzerl

and

Finlan

d

Belgium

Sweden

German

y

Canad

aFran

ce

Austria Ita

lyW

ales

Irelan

d

Englan

dRuss

ia

Scotla

nd

Fathers Sons

Note: All of these figures use our baseline linked samples, classifying immigrant families using the father’sbirthplace. Each panel plots the country-of-origin coefficients from equation (1), using alternative income measuresfor the fathers and sons. Panel (a) keeps predicted income for non-farming occupations, but adjust farmers’ incomeusing county-level measures derived from the 1900 Census of Agriculture. Panel (b) drops fathers and sons who arefarmers from the sample. Panel (c) keeps the farmer adjustments from panel (a), and also adjusts non-farmerincome using the average earnings in a person’s occupation from the 1901 Cost of Living Survey. Panel (d) adjustsson’s predicted income to incorporate their father’s country of origin into the prediction. Panel (e) adjusts father’spredicted income to not incorporate their country of origin in the prediction. Panel (f) uses the 1950 IPUMSoccscore measure.

Robustness to Sample Construction

Figure A25: Rank-rank Correlations Using Mother’s Birthplace

(a) 1880–1910 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born mother Immigrant mother

(b) 1910–1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born mother Immigrant mother

Note: This figure uses mother’s birthplace to classify men as sons of immigrants. Sons are assigned percentileranks relative to all other respondents born in the same birth year. Fathers are also ranked relative to all fatherswith children in the same birth year. The figures plot the mean income rank of children by father’s income rank, forsons with and without foreign-born mothers, as well as the corresponding regression lines.

Figure A26: Average Income Rank for Children Born to 25th and 75th Percentile, by Cohort andMother’s Birthplace

(a) 1880–1910 Cohort: 25th Percentile

2545

65

Ave

rage

inco

me

rank

Belgium

Norway U.S.

German

y

Switzerl

and

Austria

SwedenFran

ce

Denmark

Finlan

d

Englan

dW

ales

Scotla

nd

Canad

aRuss

ia

Irelan

dIta

ly

Portug

al

(b) 1910–1940 Cohort: 25th Percentile

2545

65

Ave

rage

inco

me

rank

Norway U.S.

German

y

Switzerl

and

Denmark

Finlan

d

Austria

Sweden

Belgium

Canad

aFran

ceW

alesRuss

ia

Portug

al

Englan

d

Scotla

nd ItalyIre

land

(c) 1880–1910 Cohort: 75th Percentile

3555

75

Ave

rage

inco

me

rank

Belgium

Norway

Switzerl

andW

ales

Portug

al

Sweden

Denmark U.S.

German

y

Canad

aFran

ce

Austria

Scotla

nd

Englan

dIre

land

Italy

Finlan

dRuss

ia

(d) 1910–1940 Cohort: 75th Percentile

3555

75

Ave

rage

inco

me

rank

Portug

al

Finlan

d

Belgium U.S.

Norway

Canad

a

Switzerl

and

German

y

Denmark

Wale

sIta

ly

SwedenFran

ce

Englan

d

Scotla

ndIre

land

Austria

Russia

Note: This figure plots the average income rank for children born to the 25th and 75th percentile of the parentincome distribution. Individuals are classified as sons of immigrants using their mother’s birthplace.

Figure A27: Rank-rank Correlations for Sons with Immigrant Parents

(a) 1880–1910 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born mother or father Immigrant mother and father

(b) 1910–1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

US-born mother or father Immigrant mother and father

Note: This figure classifies individuals as immigrants if both their mother and father were not born in the UnitedStates. Sons are assigned percentile ranks relative to all other respondents born in the same birth year. Fathers arealso ranked relative to all fathers with children in the same birth year. The figures plot the mean income rank ofchildren by father’s income rank, for sons with and without foreign-born parents, as well as the correspondingregression lines.

Figure A28: Rank-rank Correlations, Using Mother’s and Father’s Birthplace

(a) 1880–1910 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

Immigrant mother only, 3.12 Immigrant father only, 7.11 Immigrant parents, 20.95 US-born parents, 68.83

(b) 1910–1940 Cohort

3040

5060

70

Son

rank

0 20 40 60 80 100

Father rank

Immigrant mother only, 4.23 Immigrant father only, 6.71 Immigrant parents, 17.47 US-born parents, 71.58

Note: This figure uses mother’s and father’s birthplace to classify sons into groups. The numbers in the legendrepresent the share of the sample that falls into that group. Sons are assigned percentile ranks relative to all otherrespondents born in the same birth year. Fathers are also ranked relative to all fathers with children in the samebirth year. The figures plot the mean income rank of children by father’s income rank as well as the correspondingregression lines.

Figure A29: Intergenerational Mobility of Immigrants and the US born, Rank-Rank Correlationsby Race

(a) White immigrants, white natives

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born mother Immigrant mother

(b) All immigrants, white natives

3040

5060

70

Son

rank

0 20 40 60 80 100

Household rank

US-born mother Immigrant mother

Note: Data come from Chetty et al. (2018b). Sons are assigned percentile ranks relative to all other respondentsborn in the same birth year. Parents are also ranked relative to all fathers with children in the same birth year. Thefigures plot the mean income rank of children by their household’s income rank, for sons with and withoutforeign-born mothers, as well as the corresponding regression lines.

Construction of Linked Samples: Additional Tables

Table A1: Sample Size, by Cohort

1880–1910 1910–1940

White men, age 0–16 8,999,129 14,572,217

Linked men 2,035,587 4,292,642

living with father 1,831,450 3,897,409

white father, age 30–50 1,346,687 2,915,989

US-born or immigrant father 1,329,068 2,871,182

non-missing labor market outcomes 1,241,588 2,710,918

Share of white men age 0–16 in final sample 13.80 18.60

Note: The first row refers to the number of white men ages 0–16 in the earlier Census. Rows 2 through 6 show thesample size as we restrict the sample based on the fathers’ and sons’ characteristics. “Immigrant father” refers tosons whose fathers were born in one of the 17 source countries in our sample.

Table A2: Sample Size and Match Rate, by Father’s Country of Origin

1880–1910 cohort 1910–1940 cohort

Sample Match Rate Sample Match Rate

US-born father 891,461 24.6 2,053,821 31.9

Immigrant father 350,127 22.6 657,097 27.1

Canada 25,647 27.7 65,912 29.9

Denmark 2,820 21.8 11,961 27.0

Finland 67 17.2 5,976 21.9

Norway 7,959 19.4 27,164 31.1

Sweden 6,917 20.8 41,113 28.8

England 37,383 24.5 47,018 32.3

Scotland 8,634 21.8 13,301 30.9

Wales 3,443 15.9 3,703 23.9

Ireland 86,761 18.1 49,457 23.9

Belgium 1,059 22.1 2,387 24.4

France 5,918 23.6 4,207 28.7

Switzerland 6,010 26.0 8,764 34.4

Italy 1,112 19.1 57,798 21.4

Portugal 320 21.4 3,437 23.5

Austria 1,970 24.3 56,951 23.4

Germany 153,084 25.3 178,880 32.3

Russia 1,023 26.1 79,068 21.2

Note: This table shows the sample size of the linked samples as well as the share of white men ages 0–16 in theearlier Census who can be linked to the later Census using our baseline linking procedure, calculated separately bythe father’s country of origin.

Table A3: Comparison of Cross-Sectional and Linked Sample of Sons

(a) Sons of US-born Fathers

1880-1910 cohort 1910-1940 cohort

Non-Linked Linked Non-Linked Linked

Age 37.11 37.27 36.28 37.44

Farmer 0.24 0.25 0.30 0.33

White-collar 0.32 0.33 0.14 0.14

Skilled 0.25 0.25 0.32 0.34

Unskilled 0.19 0.17 0.24 0.19

Income 1,048.19 1,079.86 1,066.10 1,142.46

South 0.35 0.29 0.33 0.31

(b) Sons of Immigrant Fathers

1880-1910 cohort 1910-1940 cohort

Non-Linked Linked Non-Linked Linked

Immigrant (full sample) 0.26 0.27 0.22 0.24

Age 37.28 37.44 36.10 38.09

Farmer 0.27 0.28 0.36 0.37

White-collar 0.19 0.22 0.08 0.10

Skilled 0.34 0.32 0.35 0.36

Unskilled 0.20 0.18 0.21 0.17

Income 1,165.81 1,170.74 1,206.34 1,261.74

South 0.08 0.08 0.05 0.06

Note: In this table, the linked sample has not yet been restricted based on the fathers’ characteristics (i.e., whether thefather is white, aged 30–50, living with the child in the early Census, and without missing labor market outcomes).

Additional Results

Table A4: Intergenerational Mobility Estimates, by Cohort and Country of Origin

1880–1910 Cohort 1910–1940 Cohort

Intercept Slope Intercept Slope

US-born father 31.34 0.36 31.00 0.36

Immigrant father 35.68 0.32 34.67 0.37

Canada 41.52 0.23 38.97 0.27

Denmark 37.12 0.28 33.75 0.34

Finland 43.55 0.24 37.24 0.24

Norway 31.16 0.28 27.42 0.41

Sweden 33.91 0.32 34.33 0.35

England 36.82 0.32 41.08 0.28

Scotland 38.66 0.29 41.42 0.28

Wales 40.21 0.24 40.69 0.26

Ireland 42.63 0.25 45.47 0.24

Belgium 32.83 0.24 36.98 0.27

France 34.55 0.32 38.24 0.31

Switzerland 33.79 0.31 32.14 0.36

Italy 53.41 0.14 48.57 0.17

Portugal 57.83 0.00 47.49 0.11

Austria 34.38 0.39 33.25 0.42

Germany 31.56 0.37 28.09 0.43

Russia 37.95 0.43 38.15 0.44

Note: This table reports the slope and intercept from regressions of son’s rank regressed on father’s rank, by cohort andfather’s country of origin.

Robustness to Linking Procedure

Table A5: Comparison of Rank-Rank Slopes and Intercepts, by Linking Algorithm

(a) 1880–1910 Cohort

US-born father Immigrant father

Intercept Slope Intercept Slope

Baseline 31.34 0.36 35.68 0.32

Exact names 30.97 0.37 35.89 0.33

5-year band 30.29 0.38 33.90 0.35

Exact names, 5-year band 30.23 0.38 33.99 0.36

(b) 1910–1940 Cohort

US-born father Immigrant father

Intercept Slope Intercept Slope

Baseline 31.00 0.36 34.67 0.37

Exact names 31.02 0.36 34.17 0.38

5-year band 30.40 0.37 33.75 0.39

Exact names, 5-year band 30.59 0.37 33.04 0.40

Note: This table reports the slope and intercept from regressions of son’s rank on father’s rank, using samples derived using alternative linking algorithms.

Robustness to Alternative Income Measures

Table A6: 1880–1910 Cohort: Rank-Rank Slopes and Intercept, by Income Measure and Immigrant Status

US-born father Immigrant father

Intercept Slope Intercept Slope

Baseline 31.34 0.36 35.68 0.32

1900 Census of Ag. 32.94 0.32 39.74 0.26

1901 Survey, 1900 Census of Ag. 33.81 0.30 41.94 0.22

No farmers 36.34 0.27 39.32 0.22

No immigrant penalty 31.01 0.36 38.96 0.28

2nd generation penalty 30.21 0.33 28.62 0.38

IPUMS occscore 33.44 0.32 33.73 0.35

Note: This table shows the slope and intercept of the rank-rank regressions using alternative income measures. The first measure adjusts farmers’ income using the 1900 Census ofAgriculture. The second adjusts non-farmer income using the average earnings in a person’s occupation from the 1901 Cost of Living Survey. The third drops farmer fathers and sonsfrom the sample and re-runs the specification using ranks from the non-farming income distribution. The fourth adjusts father’s predicted income to not incorporate the country oforigin into the prediction. The fifth adjusts son’s income to incorporate the father’s country of origin into the prediction. The sixth uses the 1950 IPUMS occscore measure.

Table A7: 1910–1940 Cohort: Rank-Rank Slopes and Intercept, by Income Measure and and Immigrant Status

US-born father Immigrant father

Intercept Slope Intercept Slope

Baseline 31.00 0.36 34.67 0.37

1900 Census of Ag. 33.85 0.30 42.92 0.23

1901 Survey, 1900 Census of Ag. 39.02 0.19 54.49 0.02

No farmers 35.41 0.27 40.05 0.24

No immigrant penalty 30.91 0.36 37.67 0.32

2nd generation penalty 30.86 0.36 31.84 0.43

IPUMS occscore 35.15 0.29 35.38 0.30

Note: This table shows the slope and intercept of the rank-rank regressions using alternative income measures. The first measure adjusts farmers’ income using the 1900 Census ofAgriculture. The second adjusts non-farmer income using the average earnings in a person’s occupation from the 1901 Cost of Living Survey. The third drops farmer fathers and sonsfrom the sample and re-runs the specification using ranks from the non-farming income distribution. The fourth adjusts father’s predicted income to not incorporate the country oforigin into the prediction. The fifth adjusts son’s income to incorporate the father’s country of origin into the prediction. The sixth uses the 1950 IPUMS occscore measure.